MapSmith
OfficialMapSmith is an MCP server for validated, provenance-tracked geospatial analysis over vector and raster data.
Inspect datasets: CRS, geometry types, schema, extent, feature counts, raster bands, nodata, statistics, and multi-layer containers.
Run vector geoprocessing: buffer, clip, overlay, dissolve, nearest join, explode, area measurement, merge, simplify, centroid, reproject, format conversion, spatial join, and spatial SQL.
Run raster/terrain/hydrology operations: zonal statistics, hillshade, slope, aspect, flow accumulation, and watershed delineation.
Statically validate multi-step typed plans before execution, including argument, forward-reference, input-file, output-collision, and CRS-flow checks.
Execute validated plans step by step with per-step provenance manifests and a plan-level manifest.
Discover catalog operations via
list_operationsusing input/output/facet filters, then run any operation by name withrun_operation.Retrieve provenance for an output dataset and recover full analysis lineage via the
get_lineagecatalog operation.Preview vector and raster datasets in an in-chat map panel with per-layer provenance and verification status.
Surface warnings, repairs, CRS decisions, and verification results rather than silently returning suspicious results.
Report server version, licensing, and available engines.
Provides spatial SQL execution using DuckDB's dialect, including ST_* functions and file reading via read_parquet/ST_Read, enabling SQL-based geoprocessing and materialization of results as GeoParquet with provenance.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@MapSmith find parcels within 300m of rivers in parcels.gpkg and rivers.gpkg, with provenance"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
MapSmith
Ask a question about your data. Get an analysis — and the record that proves it.
Most GIS servers hand an agent a tool to run. MapSmith takes the question, works out the analysis — usually several operations, in an order that has to be right — validates the plan before a single file is touched, runs it, and leaves a record of every step — which bytes went in, with which parameters and CRS decisions, and which checks ran — that somebody else can read afterwards and rerun to confirm.
Try it
Point any MCP client at MapSmith (one JSON block) and ask, in your own words:
Which parcels lie within 1.5 km of the river and sit at 120 m or lower? Give me the mean elevation and the ground area of each.
That is five operations, two coordinate systems and forty-eight checks. What actually happens is below — the plan, the step that gets rejected for reading something a later step produces, the CRS decision behind every metric step, and an answer you can work out on paper before MapSmith sees the files. That rejection is structural, and so is every other one: a plan that is well formed and answers the wrong question runs instead, which is measured and written down further down this page.
The operations are executed by GeoPandas, DuckDB Spatial, exactextract and Whitebox Workflows — never by the model that asked for them. Every dataset lands on disk next to a lineage manifest: inputs with checksums, the exact parameters, the CRS decisions and why, engine versions, and the deterministic checks that ran on the result.
mapsmith.dev — a real terrain analysis and the manifest that came with it. Both are build products: the figure is rendered from GeoTIFFs MapSmith writes, so the page cannot drift from what the software does.
The manifest is a specified format, not
MapSmith's private output: JSON Schema, a toolchain-free validator, a conformance suite, and a
hundred-line emitter that never imports MapSmith. Records carry spec_version, and CI validates
real MapSmith output against the spec's own validator. The specification is archived and citable
as 10.5281/zenodo.22205213, a concept DOI that
resolves to the latest archived draft. Records written by 0.8.0 to 0.12.2 declare 1.0.0-draft.10, and
0.7.x's declare 1.0.0-draft.8; if the DOI shows an earlier one, the archive has not caught up, and the
repository holds the text they are written against. The names MapSmith puts
in a record beyond the ones the specification defines — every extension check name, and every
extension field in crs_decisions with the reason it is not a synonym of a key the specification
already has — are listed in docs/manifest-vocabulary.md,
generated from the source rather than maintained by hand.
Evidence before promises. A correctness suite in its own organisation, Argleton, grades MapSmith on thirty-three traps whose answers are computed on paper before any system runs: 0.00 silent errors, nothing skipped, against 0.8788 for the obvious way of writing the same code. The runs from 2026-08-30 to 2026-09-15 said 0.00 too, and on one trap they were wrong: its truth was wrong in MapSmith's favour for 26 days, the error was ours twice over, and the erratum says so. Every run since 2026-09-26, the current one included, is scored against the corrected truth. Getting here cost seven defects the suite sent back, and they are listed. Alongside it: an A/B on GABench whose headline is a null result — with the analysis that took our own positive number apart — notebooks on a real USGS DEM of Mount St. Helens, an in-chat map panel that shows the verification status of every layer it draws, and a measurement of our own tool discovery that retracted two numbers this page had already published — including the one in the bullet list below.
A whole analysis, start to finish
This is the thing MapSmith is for, and it is not a single tool call. One question — six parcels, a river, an elevation grid — becomes a plan of five operations chosen out of 81, validated before anything runs, executed step by step, and recorded: the search that narrowed the catalogue, the arguments that mattered, the CRS decision behind each metric step, and the checks that ran on every result.
Nothing in this section is drawn. benchmarks/worked_example.py builds fixtures whose answer can
be worked out on paper, asks the catalogue in the words of the problem, validates and runs the
plan, reads the manifests, and writes what follows; tests/test_worked_example.py fails if this
page and that script disagree. The position column is BM25's rather than the default engine's,
because a published figure should not depend on whether a model download succeeded on the machine
that built the page — the narrowing, which is the point, is identical on both. Two things worth watching: the middle column, where the catalogue
goes from 81 operations to a handful the caller can read; and the CRS column, where every
metric operation says which coordinate system it moved the data into and why.
flowchart TB
ASK["<b>Parcels within 1.5 km of the river whose mean ground elevation is at most 120 m, with the elevation and the ground area of each</b>"]
ASK --> PLAN{{"plan validated<br/>before anything runs"}}
PLAN -. "rejected: FORWARD_REFERENCE" .-> BAD["'mask_path' references '$buffer' which runs later — move step 'buffer' before 'near'"]
BAD:::bad
BUFFER["<b>buffer_layer</b><br/>81 operations → 28 candidates → chosen<br/>CRS EPSG:32610<br/>9/9 checks"]
PLAN --> BUFFER
NEAR["<b>clip_layer</b><br/>81 operations → 13 candidates → chosen<br/>CRS EPSG:4326<br/>12/12 checks"]
BUFFER --> NEAR
HEIGHT["<b>zonal_statistics</b><br/>81 operations → 2 candidates → chosen<br/>CRS EPSG:4326<br/>7/7 checks"]
NEAR --> HEIGHT
AREA["<b>measure_area</b><br/>81 operations → 25 candidates → chosen<br/>CRS WGS 84 (ellipsoidal)<br/>10/10 checks"]
HEIGHT --> AREA
FILTER["<b>select_features</b><br/>81 operations → 25 candidates → chosen<br/>CRS EPSG:4326<br/>10/10 checks"]
AREA --> FILTER
OUT[["3 parcels, each with elevation and ground area"]]
FILTER --> OUT
classDef bad stroke-dasharray: 4 3what the agent asks for | it declares | candidates | picked | at position |
“everything within one and a half kilometres of the river” | vector, dataset:vector, 1 dataset(s), line | 28 of 81 |
| 2 |
“keep only the parcels that fall inside that strip” | vector, dataset:vector, 2 dataset(s), polygon | 13 of 81 |
| 1 |
“how high is the ground under each of these parcels” | raster, dataset:vector, 2 dataset(s), polygon | 2 of 81 |
| 2 |
“how big is each one on the ground” | vector, dataset:vector, 1 dataset(s), polygon | 25 of 81 |
| 1 |
“drop the ones where the ground is above 120 metres” | vector, dataset:vector, 1 dataset(s), polygon | 25 of 81 |
| 1 |
step | operation | arguments that mattered | CRS decision, recorded | checks |
buffer |
|
|
| 9/9 |
near |
|
|
| 12/12 |
height |
|
|
| 7/7 |
area |
|
|
| 10/10 |
filter |
|
|
| 10/10 |
Every step is inside the plan, the last one included: select_features took 4 rows and returned 3, with a manifest like every other write. This step used to run outside the plan, because the only operation that could answer it was run_sql — which takes its inputs inside a SQL string, declares zero datasets, and therefore cannot join the plan's dataflow. That boundary is deliberate and has not moved: substituting $step into arbitrary strings would be a grammar in which a planner assembles a path out of text. What changed is that it is no longer the only way to ask.
The answer, which can be worked out on paper before MapSmith sees the files: the parcels are squares of 0.0015° at 46.2°N, so each is about 119 m by 167 m, and the elevation ramps west to east across the fixture.
name | mean | min | area_m2 |
North Field | 104.85 | 104.14 | 19303.33 |
Mill Meadow | 110.51 | 109.8 | 19303.33 |
Old Orchard | 117.58 | 116.87 | 19303.33 |
The rejected plan is the honest half. Steps in the wrong order are the dominant failure class in
the agent benchmark, so the example includes one and shows what the validator says about it,
before any file is touched. It earned that place while this was being written: the first version
of the plan passed distance_m where the operation declares distance_meters, and the validator
named the argument and listed the three it accepts.
You can step through this on mapsmith.dev, one beat at a time, along with a second case that GitHub cannot show as well: a file MapSmith refuses. Both are executed while that page is built. The residue of the second one is here, because it is the half nobody demonstrates.
What it declines to answer. Asked “How high is the ground at each of these two wells?” against a GeoTIFF that opens in every viewer:
describing the file first | |
| sidecar (.aux.xml) |
| terrain.tif.aux.xml |
| cell 10 x 10 at (500000, 5030000) |
terrain.tif is georeferenced twice and nobody chose: the GeoTIFF's own tags say cell 10 x 10 at (500000, 5030000), and terrain.tif.aux.xml beside it says something else. GDAL prefers the sidecar, which is correct — that is how an override works — but sample_raster_at_points would then report numbers from a file you did not name, and this record could not say which. describe_dataset lists both. To choose, either remove the sidecar or set GDAL_GEOREF_SOURCES=INTERNAL for a run that must use the file's own.A number would have come back. It would have been read off a grid a hundred kilometres away and at twice the cell size, with nothing in the record able to say which of the two readings produced it.
And the analysis comes back out of the file. Given only the last output of the five-step run above — no plan, no filenames, no field pointing at another record — the chain is recovered by hashing it and following every input digest back:
5 operations recovered back to 3 original dataset(s), every one passing its critical checks.
<- select_features 10 checks
<- measure_area 10 checks
<- zonal_statistics 7 checks
<- clip_layer 12 checks
<- buffer_layer 9 checks
(original) elevation.tif
(original) parcels.gpkg
(original) river.gpkgRelated MCP server: GIS MCP Server
Quickstart
Add MapSmith to any MCP client over stdio (Claude Desktop, Claude Code, Cursor, VS Code):
{
"mcpServers": {
"mapsmith": {
"command": "uvx",
"args": ["mapsmith"]
}
}
}Docker is the supported path, and confines the server to the directory you mount:
{
"mcpServers": {
"mapsmith": {
"command": "docker",
"args": ["run", "-i", "--rm",
"-v", "/absolute/path/to/your/data:/data",
"-e", "MAPSMITH_WORKSPACE=/data",
"ghcr.io/mapsmith-ai/mapsmith"]
}
}
}One-click installs:
or from a terminal: code --add-mcp '{"name":"mapsmith","command":"uvx","args":["mapsmith"]}'
To check it runs before wiring a client, uvx mapsmith starts the server on stdio
(Ctrl-C to quit) — it speaks MCP, not a CLI, so a silent prompt means it is working.
This page describes 0.12.2, which is what that command installs. Its
release notes add benchmark evidence and refreshed
discovery labels. When main runs ahead of
the published artifact this paragraph says so and names the difference — a reader should never
have to find out by calling a tool that is not there. New in 0.12.1
(release notes): semantic verification of every primary,
secondary and native geometry CRS declaration in calculate_field. New in 0.12.0
(release notes): restricted numeric and boolean field
calculation, executable typed-DAG input contracts and corrected multi-step method-choice
benchmarks. New in 0.11.0
(release notes): the validated project brief, bounded agent
repair, vector/raster conversion and QGIS native:buffer, taking
the catalogue to 80 operations behind 29 tools. If you run 0.10.0 or earlier under
MAPSMITH_WORKSPACE, update: a coordinate reference system -- a tool argument, or the one a
file declares -- could make PROJ read a file outside the workspace or open a network share, and
in run_sql DuckDB's own PROJ could do the same; 0.10.0 closed that for tool arguments and for
GeoPackage, shapefile and GeoParquet inputs, and 0.10.1 for every other format MapSmith accepts
-- under a workspace it now reads each input's coordinate system before GDAL opens the file,
and refuses a format it does not recognise. SECURITY.md has all of it. From
0.9.0, 0.10.0 also: every operation that brings a vector input onto another CRS -- the eight
that align a second input, and the samplers, zones, masks, pour points and stations of the
raster ones -- chooses the datum transformation the way reproject_layer does, which moves
their results where PROJ's own pick was the ballpark and a covering operation or a Greenwich
twin exists (Monte Mario from Rome: up to about 80 m); a datum shift a CRS itself declares
(+towgs84) is no longer recorded as a ballpark; nearest_neighbour_index works in linear
memory and refuses a feature with no position; under a workspace run_sql runs only SELECT,
with coordinate systems and dataset paths as literals; and 103 catalogue phrasings from the QGIS
Training Manual's exercises moved the discovery figures on this page. From 0.8.0, 0.9.0 adds
apportion_by_area, which moves counts and rates from the zones they are reported in to zones
that cross them, and makes a record's transformation describe what PROJ did to the data's own
coordinates rather than to a point chosen from the CRS. That changes what reproject_layer
writes in two cases: coordinates outside every published operation's area are no longer
moved by an operation from elsewhere, and a datum counted from another meridian whose
operations are published only for its Greenwich twin -- Monte Mario from Rome, EPSG:4806 --
goes through that twin. The changelog names both. If you run 0.4.0 to 0.6.1, 0.7.0 and
later change answers you have: every position read from a point-registered raster
(AREA_OR_POINT=Point, the USGS and Copernicus DEMs) was half a cell off, and the
changelog says which operations and why. From 0.7.x, 0.8.0 adds
cluster_points_by_distance and a geometry facet for the catalogue search, digests a
shapefile with its attributes and its CRS, reads any single-layer file geodatabase, and fixes
operations that accepted the wrong geometry and answered anyway (zonal_statistics on line
zones returned their length as a pixel count).
If you read manifests, 0.10.0 renames no key: records still declare 1.0.0-draft.10.
Two keys of MapSmith's own are new inside crs_decisions.transformation:
x-mapsmith:shift_from_crs_definition, when the shift is written into the CRS itself, and
x-mapsmith:default_was_ballpark, when the engine's own choice would have applied no shift.
If you read manifests, 0.9.0 renames no key: records still declare 1.0.0-draft.10.
Three keys of MapSmith's own are new: crs_decisions.x-mapsmith:areas_measured_on, from
apportion_by_area, and x-mapsmith:ballpark_share and x-mapsmith:outside_area_of_use
inside transformation; so are three check names, all from apportion_by_area. A run that
fails where an earlier run's dataset still sits writes its record beside the earlier one, as
<output>.failed.provenance.json, instead of replacing it.
If you read manifests, 0.8.0 renames one key: records declare 1.0.0-draft.10, and a
second input's own georeferencing facts, inputs[].x-mapsmith:environment on 0.7.x, are
inputs[].environment, the name the specification gives them since draft.10. Each input of
an operation that reads more than one dataset also says which argument read it
(inputs[].argument).
If you read manifests, 0.7.x renames no key: records declare spec_version
1.0.0-draft.8, two drafts past 0.6.0's draft.6, and both drafts only narrowed where a
producer may add a key of its own — which MapSmith already did the narrower way. From 0.5.x,
0.6.0 renamed x-mapsmith:round_trip to round_trip; from 0.4.0, 0.5.0 had renamed eight
crs_decisions keys, the top-level mapsmith_version (read producer.version), two fields of
the round trip, two of repairs[] and engine.geometry_library. The changelog lists each with
what to read instead.
Then ask your agent things like:
"Take parcels.gpkg, keep only the parcels within 300 m of the river in rivers.gpkg, and give me the result with the analysis lineage."
The Docker image includes the [raster] and [whitebox] extras. With uvx, pick your
own: uvx --from "mapsmith[raster,whitebox]" mapsmith. Docker — or uvx on a machine
with working wheels — is the only supported installation path: geospatial native
dependencies across three OSes are a support black hole, and issues about broken local
environments will be redirected here.
Two things about the image, because they change what happens on your machine: it sets
MAPSMITH_WORKSPACE=/data itself (the -e above is explicit, not required) and runs as
uid 1000, so pass --user $(id -u):$(id -g) if the directory you mount belongs to another
user; and it is built for amd64 only, so on Apple Silicon it runs under emulation.
What you get back
Every dataset comes with the file below, written next to it as
<output>.provenance.json — enough to re-run the analysis without the model that asked
for it:
{
"spec_version": "1.0.0-draft.10",
"producer": {"name": "mapsmith", "version": "0.12.2"},
"operation": "buffer_layer",
"parameters": {"distance_meters": 300.0},
"inputs": [{
"path": "rivers.gpkg",
"sha256": "b24b884f49eee431133d443d842557d3ed21f3978e2c4b1e8e072fd1240effe8",
"crs": "EPSG:4326"
}],
"crs_decisions": {
"analysis_crs": "EPSG:32632",
"reason": "estimated UTM zone for metric buffering on a geographic CRS",
"round_trip": {
"transformation": {
"pipeline": "+proj=pipeline +step +proj=unitconvert +xy_in=deg +xy_out=rad +step +proj=utm +zone=32 +ellps=WGS84",
"accuracy_m": 0.0,
"is_ballpark": false
},
"return_transformation": {
"pipeline": "+proj=pipeline +step +inv +proj=utm +zone=32 +ellps=WGS84 +step +proj=unitconvert +xy_in=rad +xy_out=deg",
"accuracy_m": 0.0,
"is_ballpark": false
}
}
},
"engine": {"name": "geopandas", "version": "1.0.1"},
"environment": {},
"verification": [
{"name": "input_crs_present", "passed": true, "detail": "'input_path': EPSG:4326"},
{"name": "input_not_empty", "passed": true, "detail": "'input_path': 1 features"},
{"name": "result_not_empty", "passed": true, "detail": "1 features"},
{"name": "crs_present", "passed": true, "detail": "EPSG:4326"},
{"name": "crs_matches", "passed": true, "detail": "expected EPSG:4326, got EPSG:4326"},
{"name": "geometry_valid", "passed": true, "detail": "all valid"},
{"name": "x-mapsmith:no_geometry_is_empty", "passed": true, "detail": "none empty"},
{"name": "feature_count_exact", "passed": true, "detail": "expected 1, got 1"},
{"name": "geometry_types", "passed": true, "detail": "expected subset of ['MultiPolygon', 'Polygon'], got ['Polygon']"}
],
"started_at": "2026-08-18T10:15:03Z",
"finished_at": "2026-08-18T10:15:04Z"
}The checksum and the timestamps are made up; everything else — the CRS decision, the
transformation, every check name and detail — is copied from a record buffer_layer emits
for exactly this call. The real file also carries output, repairs, notes and a
critical flag on each check, left out here for length.
tests/test_showcase.py validates this block against
the specification's own validator and re-runs the operation to compare its crs_decisions
keys with the emitted ones, because a record typed by hand drifts silently while every
generated surface stays correct: it lacked spec_version for two releases, and on
2026-09-06 it was still recording a round trip without the
key that says the coordinates came back — x-mapsmith:round_trip then, and
round_trip
in the specification since 1.0.0-draft.6. This is the record a third-party implementer copies.
Trimmed for the page, not for the file: the real record also carries the output's own path
and hash, any geometry MapSmith had to repair, and the notes it made about how the inputs
were handled — and every check says whether it was critical and, when it failed, what to do
about it. get_provenance returns it for any output — one operation, in full.
For the analysis rather than the step, run_operation with get_lineage takes the final file, hashes it, finds the record claiming those exact bytes and follows every input digest back to the originals. No field points at another record: the chain is recovered from content, so it survives a rename and breaks only when the data changes. It reports a step that failed its checks as one, and says where the walk stopped rather than presenting a history it did not reach.
Why MapSmith
The unit of work is the analysis, not the tool call. Other GIS servers expose operations and let the caller fire them one at a time; a question that takes five steps is five unrelated calls, and nothing above them knows they belong together. MapSmith takes a typed plan, refuses it if a step reads something a later step produces, runs it in order, writes a manifest for each step and a plan-level record tying them together, and leaves the chain recoverable afterwards from the content digests alone. That is the difference a wrapper around a toolbox cannot have, because there is nothing above the single call to put it in.
Real geoprocessing, not map CRUD. Built on the proven open geospatial stack: GDAL, GeoPandas, Shapely, DuckDB Spatial, Whitebox Workflows and exactextract ship today and QGIS Processing's
native:bufferthrough a GPL-isolated sidecar.Provenance by design. Every layer MapSmith produces ships with a machine-readable lineage manifest — source datasets with checksums, tools executed, exact parameters, CRS decisions, software versions, timestamps. Everything needed to re-run the analysis without the LLM is in there. No AI slop.
The engines compute, the model orchestrates. Geometry and numbers only ever come from deterministic tool executions — never from model output.
81 operations, reached through 29 goal-level tools. The catalogue is the breadth and the tool list is the choice; it is searchable rather than dumped, because tool-selection accuracy degrades once a few dozen tools are exposed at once, and fastest when two of them apply to the same input. Capability count has no such ceiling, so capability lives in the catalog. Search narrows it on what you declare and then hands over what survives rather than ranking it for you, because measurement said ranking is the wrong verb — see Finding the right operation.
Model-agnostic infrastructure. Claude, GPT, Qwen, Kimi, GLM — anything that speaks MCP, cloud or local. The leverage is better contracts (typed plans, actionable error codes, a searchable catalog), not weights we would have to maintain. See the manifesto.
Tools
Tool | What it does |
| CRS, schema/bands, extent, nodata and statistics of any vector or raster dataset |
| Metric buffer with automatic UTM estimation for geographic CRS |
| Clip a layer with a mask layer |
| Set-theoretic overlay (intersection/union/difference/…); dropped lower-dimension pieces are declared in the manifest |
| Merge features per key; the aggregation is recorded in the manifest and the group count verified |
| Nearest neighbour with the distance in meters, UTM-measured on geographic CRS (decision recorded) |
| Multi-part to single-part, with the part count verified in closed form |
| Area in m², always: ground on the ellipsoid, or planar converted with the CRS's own declared linear unit (survey feet are not metres). Invalid rings repaired before measuring, and a plane that is not equal-area here comes back with the ratio against the ground area |
| Append layers (schema union); null-filled columns are named in the manifest, the count verified against the sum |
| Douglas-Peucker with the drift measured: area/length before and after recorded in the manifest |
| Geometric centroids computed in a metric CRS, never on degrees (decision recorded) |
| Convert between GeoParquet/GeoPackage/GeoJSON by output extension, re-read and verified (count and CRS). Two conversions are refused with the reason rather than performed: shapefile output, which truncates field names to 10 characters silently, and GeoJSON for a non-WGS84 layer |
| Reproject to any CRS (EPSG code or WKT) |
| Join by spatial predicate, auto-routed to the fastest engine (SedonaDB > DuckDB > GeoPandas) |
| Spatial SQL (DuckDB dialect) over GeoParquet and GDAL formats |
| Raster statistics per vector zone with exact fractional pixel coverage ( |
| Shaded relief from a DEM, in-memory Whitebox engine ( |
| Slope gradient from a DEM in degrees, percent or radians; geographic-CRS DEMs refused ( |
| Downslope azimuth from a DEM, 0 = north; flat cells are −1, not nodata ( |
| D8 flow accumulation with automatic depression filling ( |
| Watershed delineation from a DEM and pour points ( |
| Interactive in-chat map (MCP Apps) of any datasets, with a provenance card and verification status per layer |
| Statically validate a multi-step plan before running anything: operations, arguments, references, input files, simulated CRS flow |
| Render a project brief only after plan validation, with a declared source for every statement |
| Validate then run a plan step by step, with per-step provenance and a plan-level manifest |
| Return the manifest of any MapSmith output: one operation, its inputs and their digests, and what was checked |
| Catalog search: narrows on what you declare, then returns the surviving set to choose from ( |
| Run any catalog operation by name, including those with no tool of their own; arguments validated against the catalog before anything runs |
| Version, license, available engines |
Finding the right operation
Those are the tools an agent chooses between. Behind them the catalog holds every
operation MapSmith can perform — 81 today, and 56 of them have no tool of their own — and
it is built to hold thousands. (One of the 81 is marked planned and says so when asked:
the roadmap is in the catalog on purpose, so an agent can answer "not yet" instead of
inventing a call.)
The split is the design: tool-selection accuracy degrades past a few dozen exposed tools, while capability count has no such ceiling. That makes reaching scale a retrieval problem, so it is treated as one — and measured like one.
First it narrows, deterministically, on things the caller already knows. Every entry
declares what data it takes (vector, raster, dataset, plan, none), what it hands back
(dataset:vector, dataset:raster, answer, description), whether it demands a projected
CRS, and which family it belongs to.
Measured over 118 answerable requests written by two other model families from job scenarios — a hydrologist with a flood report, a surveyor arguing with a field measurement — neither of which was shown this catalog, because a model handed the entry writes a paraphrase of the entry:
what the caller declares | candidates left | BM25, found@3 | embeddings, found@3 | right answer in what comes back |
nothing — words alone | 81 | 47% | 32% | 47% |
what data I have | 54 | 52% | 35% | 53% |
+ what I want back | 33 | 61% | 52% | 67% |
+ how many datasets I have | 18 | 66% | 62% | 89% |
+ what geometry it holds | 15 | 75% | 72% | 99% |
The last row is new on 2026-09-29, and the row above it shows why. Thirty-one operations
now take one vector layer to another, one more than the threshold below, so a caller who
declared everything it knew was handed a shortlist for the most common request there is, and
delivered fell from 98% to 84%. The fix is one more fact the caller holds: whether its layer
is points, lines or polygons, as a list when it is mixed. An entry declares the families it
accepts only where its code refuses the others, and a test calls each one with every family it
does not declare and requires the refusal, so the declaration cannot hide an operation from a
caller it would have worked for. Writing that test found zonal_statistics accepting line
zones and returning their length as the pixel count; it now refuses them.
Two ranking columns, and that is a correction. This table used to carry one, computed with the default engine — which is the embedding one where its model loads and BM25 where it does not. So the published figures were a measurement of what the machine could download, and a CI run that met a 429 from Hugging Face recomputed the first row as 28% where this page said 18%. Not a flaky test: a number that had never been reproducible on a machine without the model, published under a sentence promising it could be checked.
The two also differ in a way worth seeing. In this run BM25 leads at every declaration level, by fifteen points on words alone. The embedding engine earns its place on the phrasings it has never seen, not on the ranking once the set is small.
The last column reports whether the accepted answer appears in the returned payload: 117/118, or 99%. Whole-set delivery is a different measurement: 114/118, or 97%. Four sets exceed thirty candidates; three of those still return the accepted answer in the ranked fallback. Ranking decides the order, not membership.
The third row is the scaling wall, and we hit it in one afternoon. On 2026-08-29 the catalogue went from 51 operations to 61. Two rows of that table got worse: the commonest surviving set went from 26 candidates to 34, past the point where the whole set can be handed over, and delivered fell from 100% to 45% while found@3 fell from 48% to 36%. Adding capability had made discovery worse — the failure this page had predicted at eight hundred operations and met at sixty-one.
Raising the threshold would have postponed it by about ten operations. What fixed it is the fourth row: how many datasets you are holding. That is a fact about your situation — one layer or two — not a guess about our vocabulary, it is derivable from each operation's own signature so a test can check the declaration against the code, and it takes the median surviving set from 34 to 9. The catalogue grew by a fifth and discovery got better, but only because a facet arrived with it. That is the trade this design makes, stated rather than discovered later.
It happened again the next day, and this is what watching a curve is for. On 2026-08-30 the
catalogue went from 61 operations to 71. Every ranking figure in that table fell — 28% to 25%
bare, 34% to 27% on the input kind — and the delivered column of the second row fell from 48%
to 27%, because more requests now leave a set too large to hand over whole. The bottom row did
not move: 97%, the same as at 61. Ten more operations, no new facet, and the guarantee held,
which is the first time growth has been absorbed by the facets already there. (Not «and at 51»:
that bottom row is the arity facet, and dataset_inputs did not exist at 51 operations. The 100%
quoted above at that size is the row above it. Two different rows under one sentence is the kind
of comparison this page exists to refuse.)
And again at 74, with two operations that a caller is unusually likely to want. select_features and extract_layer are the remedies MapSmith's own error messages had been recommending, so they sit in the busiest corner of the facet space: the average surviving set went from 16 to 17 and the second row's delivered did not move. The bottom row held at 97% for the third catalogue size running. The margin to the wall is now 13. (Those are the figures as measured on 2026-08-31. They were recomputed on 2026-09-01 against human answers — see the note under the table — which moved them without any catalogue change: the surviving set reads 16 and the bottom row 98%. A paragraph about a transition keeps the figures of the transition.)
And at 76 the cost showed where the new operation lives. summarize_points_in_polygons takes two vector layers and returns one, which is the busiest corner of the facet space — count_in_polygons, spatial_join and clip_layer are its neighbours — so the bottom row's found@3 fell from 58% to 55% under BM25 while the average surviving set stayed at 16 and delivered held at 98%. The operation was added because «average pH by field block» took two calls to answer and so, to somebody searching, did not exist; three points of ranking at the fullest declaration is what that cost, and the guarantee that matters did not move.
And when an existing operation gained an optional input, with no operation added. zonal_statistics took a weights raster on 2026-09-25, so it reads two datasets or three, and an arity that varies is declared as none at all — the rule run_sql and merge_layers follow, because declaring 2 would hide the weighted form from a caller who correctly says they hold three. The price is that it now survives every arity: the average surviving set went from 16 to 17 and the bottom row's found@3 from 55% to 53%, with delivered still at 98%. A facet that stops discriminating for one entry costs every search a candidate.
That is the shape of the trade, and it says when the next facet is due. The figure to watch is not found@3 — a ranker will always get worse as the catalogue grows, and it is a hint. It is the average surviving set at the fullest declaration, the fourth column of that table: 9 at 51 operations, 14 at 61, 16 at 72, 16 at 74, 16 at 75, 17 at 76, 15 at 77, 15 at 78, 15 at 81. (The point at 77 is the first with the geometry declared; without it the set would be 17.) When that crosses the threshold, delivery stops being a property and starts being a ranking again, and the lasting answer is another fact the caller already knows. A bigger threshold is an answer only as far as the caller still chooses well among more, and that is measured below, not assumed. (It said median until 2026-08-29, and published the mean: the median at 81 is 13. The distribution is skewed — most requests leave a small set and a few leave a large one — so the two numbers say different things and the mean is the pessimistic one, which is the right one to watch.)
The requests, both labels and the harness are all in the repository:
tests/data/discovery_queries.json and
benchmarks/discovery_report.py, which recomputes every
number above from those files with no network and no model — so they can be checked rather than
believed, and tests/test_discovery_report.py fails if this page and the harness disagree. The
one exception is the 69%: reproducing that needs the model that did the choosing, and the report
says so where it stops. The measurement of the threshold below is the other way round: its
choices are in the repository, so its figures can be recomputed without the model.
So it hands over the set instead of picking for you. At thirty survivors or fewer list_operations
answers with status: "choose": every candidate, ordered as a hint that says it is a hint, each
carrying the sentence that separates it from its neighbours. Over those 118 requests the surviving
set has a median of 13 and exceeds 30 for four of them — 114/118, or 97% receive it whole.
The table's 117/118, or 99% includes ranked fallbacks: three of those four still return the
accepted answer. The payload is about 2,100 tokens, less than one wrong operation costs
to run and undo.
Thirty is measured, and a larger threshold was tried first. Raising it was the obvious answer
to the thirty-first vector-to-vector operation, and models get better at reading long lists, so
before moving it we measured whether one still chooses as well from more
(benchmarks/choice_at_scale.py, every choice in
benchmarks/choice_at_scale.json). Each of the 118 requests
was handed over in a set of 30, of 40 and of 50 candidates, filled with the operations the ranker
finds closest, so the extra ones are the confusable ones; twelve choosers (Claude Opus 5.5) so
that none saw a request twice; and each request compared with itself.
the larger set | pairs | right from 30 | right from the larger set | right only from 30 | right only from the larger |
40 candidates | 116 | 85.3% | 78.4% | 9 | 1 |
50 candidates | 118 | 84.7% | 79.7% | 8 | 2 |
Seven points lost at forty and five at fifty, on the same requests, so the threshold stayed and
the fix went where this section says the lasting one belongs: the geometry row above. A first run
of this measurement had said the opposite, that forty cost nothing; it compared a different mix
of requests at each size rather than each request with itself, and review caught it before it
reached main. A test holds the threshold at or below the largest size measured without loss,
so moving it means re-running the measurement, and a stronger model is a reason to re-run it,
not to edit the number.
(That sentence used to say the set had a median of 26 and never exceeded 30. It was false before this catalogue reached 72 operations and nobody noticed, because the test that checks this page against the harness read the table and not the prose around it. It reads both now.)
Three measurements say this is the right shape, and the third is the one that settles it:
our ranking puts the answer in the top three | 75% |
a model handed the same candidates and asked to choose gets its first pick right | 69% |
the two labellers who wrote the ground truth agree with each other | 75% |
These figures went up on 2026-09-01 because the measurement changed, not because the ranker did. Somebody who does this work answered all fifty of the requests the two model labellers had disagreed on, and a request can have more than one acceptable answer: two experienced analysts reach the same result with different tools. So a hit is now counted against every operation a professional would accept rather than against one label, and found@3 rose four to five points. Nothing in the ranking code changed. Read the other way round, the older figures were understating by that much — they scored a system as having failed when it returned the other defensible answer — and the honest description of this table is the answer is among the ones a professional would accept, which is a property, not an accuracy.
Where the secondary answers came from is recorded rather than smoothed over: the primary on each of those fifty is a human choice, and the secondaries were proposed by a third model and adopted wholesale rather than judged one at a time. Two different strengths of evidence, and the file says which is which.
The current model labels were written on 2026-11-04 against the catalogue of 81 operations, by
Claude Sonnet 5 and Claude Haiku 4.5. Their raw answers and the dataset and catalogue hashes are
in benchmarks/discovery_labels_2026-11-04.json. The original Claude/Gemini labels, written on
2026-08-28 against 51 operations, remain unchanged in the file, so the dated comparison is not
erased. Human answers still take precedence wherever one exists — benchmarks/ingest_answers.py
is how they get in, and every figure above is computed against a human answer where there is one.
All three are over the same 118 requests, which matters: agreement measured over all 155 requests in the file is 60%, which also counts requests both labellers called unanswerable. Quoting that 60% beside the 75% computed over the 118 would compare two populations.
The last row is a ceiling, not a baseline, and the second row sits at it rather than below it. When two competent labellers disagree one time in four about which operation answers a request, "the right one" is not a single value to rank toward, and a system scoring above that is fitting one annotator rather than getting better. Two GIS analysts with thirty years each do the same job with different tools and neither is wrong.
That is why the answer is a set and why its reason field says, in words, that the order is a
hint and that two defensible candidates are a question for the person who made the request.
The caller — an agent with the conversation in context — knows things no ranking can. Where it
does not, the human does.
The remaining honesty: the ceiling was measured between two language models. Whether human GIS analysts agree with each other more, less, or about the same is unmeasured, and until it is, these numbers are reported as agreement with model-written labels and never as accuracy.
The family is the one facet that orders instead of filtering, and that is a correction. It
used to be a hard filter like the others. It is not like the others: input kind and projected-CRS
are facts about the data in hand and output kind is what the caller wants, but family is a guess
about our taxonomy, which the caller cannot see. Measured, it removed five candidates out of
fifteen — and when the guess was wrong it removed the right operation, with no error, leaving a
confident answer assembled from neighbours. Every request in the independent set has 4.4 plausible
families. That is the silent-failure class Argleton measures in other
people's systems, sitting in our own discovery layer, so it now sorts: declaring the family lifts
it to the front and costs positions when wrong, never the answer. The hard cut stays available on
catalog.applicable, where asking for it means it.
We do not need a model to extract those facets, because the caller is one. An MCP client is an
LLM with the context we lack — it knows what file it is holding and what it is trying to produce.
So list_operations asks for them in its schema, and its description leads with why. This is the
same shape as LlamaIndex's Auto-Retrieval or LangChain's Self-Querying, minus the model those have
to host: here it is already on the other end of the protocol. A geographic raster is never offered
slope, because slope refuses one — a property of the data, checked in code, no model in the
loop.
Then it ranks, with two engines that both always run. list_operations takes engine:
auto (the default), lexical, or vector. Every result carries the engine that produced it,
because a BM25 score of 10.03 and a cosine of 0.38 are not on the same scale.
engine | what it is | what it guarantees |
| The embedding engine, falling back to BM25 when the model cannot be loaded | An answer on a machine with no network, and a field saying which engine gave it |
| Okapi BM25, ~40 lines, no model and no network ever | Identical scores on every machine; term-sorted accumulation, because float addition is not associative |
| Static embeddings — a token lookup plus pooling, no transformer, no GPU. Model revision pinned in the source, 512 dimensions, ~130 MB fetched once | Bit-identical across calls in one process (measured, multiprocessing off), with the vectors pinned by a golden-vector test — so a change in the model, the tokenizer or the pooling fails a test instead of an analysis |
The default was lexical until the measurement said otherwise, and the measurement is the interesting part. Golden queries written by whoever wrote the catalog share its vocabulary, so they test word overlap dressed as retrieval: on those, BM25 scores 100% found@1 and embeddings 60%. Re-phrased the way somebody with a problem actually phrases it — "the coastline is 400000 nodes and the browser dies" rather than "simplify the geometry" — the finding reversed, on a catalogue of fifty-one entries. It has since reversed back, and both engines degrade as the catalog grows:
catalog size | BM25 found@3 | embeddings found@3 |
10 | 77% | 83% |
30 | 62% | 60% |
81 | 45% | 45% |
This table used to say the opposite, and the reversal is the finding. Published at 10/30/51 it read 78/83, 47/65, 40/55 — embeddings ahead at every size — and the sentence under it said BM25 degrades faster, which is why the embedding engine became a dependency rather than an extra. Recomputed today the crossover has moved: embeddings still lead on ten entries, and from thirty up BM25 leads by a margin that widens with size. Part of that is the catalogue itself, because the distractors are drawn from it and it has grown from fifty-one entries to the last row — which is the point rather than a caveat. The near-neighbour effect the eight-hundred-operation test predicted has arrived in our own catalogue, and the two tables that used to disagree now agree.
The curve is recomputed by tests/test_retrieval_degradation.py and compared with this table, so
it cannot go stale again in silence — which it did for three catalogue sizes.
And that finding does not survive being scaled up — measured the same day it was published. The distractors above are drawn from our own entries, which are semantically spread out. Growing this catalog means adding near neighbours: hundreds of raster and terrain operations that resemble each other. Re-run against 800 real GIS operations, taken from a library that ships them with their own descriptions, the ranking reverses and the embedding engine degrades faster:
catalog size | BM25 found@3 | embeddings found@3 |
74 — our own entries, no foreign distractors | 50% | 40% |
200 | 48% | 25% |
800 | 35% | 20% |
Embeddings blur near neighbours; an exact term either matches or does not. These two
measurements used to disagree, and both were kept because they answered different questions:
which engine suits the catalog we have, and which survives the catalog we plan. They agree
now — the near-neighbour effect this one predicted has arrived in our own catalogue, so BM25
leads at both scales, and the second question has the answer neither. The embedding engine is
still the default, and the measurement that made it one no longer says so: that is a decision to
take rather than a number to quietly restate. At 800 entries the better engine is wrong two
times in three, so scale will not be bought by choosing a better ranker.
test_retrieval_at_scale.py keeps the projection under measurement rather than under opinion.
And the narrowing does not scale on its own either — this page claimed otherwise and was wrong. It said the facets leave sixteen candidates at 800 operations just as they do at 200. The sixteen is real and it is produced almost entirely by family: those 803 operations are all raster-in, raster-out, so input kind and output kind cut nothing at all, and only the taxonomy does — a choice among 43 families that the caller has to guess. Which is exactly the facet that must not filter.
So the open problem has a sharper shape than "ranking is hard". What is needed at a thousand
operations is more facts a caller can state without knowing our taxonomy — how many inputs an
operation takes, whether it changes geometry or only attributes, whether the output has the same
number of features as the input. Those are structural properties of the operation, they are
checkable against the code rather than declared by hand, and they separate the pairs a bag of
words cannot: spatial_join from overlay_layers, flow_accumulation from extract_streams.
That work is not done, and until it is, the honest claim is the measured one: the guarantee above
holds at eighty-one operations, not at eight hundred.
How an entry has to be written is a published specification, not a convention:
docs/catalog-entry-spec.md, with a normative
JSON Schema that every entry validates against in CI. Each field
is there because a measurement said so — including the two that measured to nothing and are
documented as such, because a spec that only reports what worked is an advertisement.
And discoverability is a contract per operation, not an average. A catalog-wide 90% found@3
over fifty entries means five are invisible and the average will not say which. So every available
entry is probed with its own first worked example, with its own facets declared
(test_discovery_contract.py, parameterised over the catalog, so a new operation is under
contract the moment it is added). Two things are required of it: the facets the entry declares
must never drop that entry, and the entry must reach the caller.
Its rank is no longer one of them, and removing that is the point. The contract used to demand the top three. That looks like a discovery contract and is a ranking contract, with one bad property: the only way to repair a failure is to reword the entry until the ranker likes it. Fifty entries tuned that way score nineteen points better on examples we wrote than on requests written by anyone else — that gap is measured, and it is where a published 70% on this page turned into 51% overnight. A test whose repair procedure is fit the text to the scorer manufactures the number it reports. What remains under contract is the part that is deterministic and ours; rank inside the delivered set is still measured, and no longer fails a build.
The old form still earned its place the first time it ran. centroid_layer advertised “label
points for a polygon layer” and ranked below point_on_surface. The ranking was right: a centroid
can fall outside its own polygon, which is Argleton trap 014 — our catalog
was recommending the defect our own suite measures. The example changed, not the score.
And when the two engines agree on nothing, the search says so instead of answering. This is
the failure that measurement turned up in our own product: asked "send an email to my
accountant", the embedding engine returned idw_interpolation with the same confidence as a
real answer — a silent error in the layer whose job is to prevent them. A similarity threshold
does not fix it, because there is no line to draw: "convert this mp4 to a gif" scores above
sixteen of twenty genuine queries. What does separate them is the two rankers landing on
nothing in common — mean top-3 overlap 0.90 of 3 when an answer exists, 0.18 when it does
not. So a query the catalog cannot place comes back as status: "unsure", carrying both
engines' guesses and the question that narrows the catalog deterministically: what kind of data
do you have. It fires on 9 of 11 unanswerable queries and suppresses 1 correct answer in 20.
And when the facets leave nothing at all, it says which declaration did it. Zero
candidates used to fall through the branch above and come back as "0 operations survive,
which is few enough to read" — prose that means nothing and, worse, an empty candidate
list, which an agent reads as MapSmith cannot do this. It was found by the discovery log
below on its first real session: "how much land is in each of these parcels" with
produces="answer" left nothing, while measure_area computes exactly that and declares
dataset:vector because it writes the areas into a column. So that case is now its own
answer — each declaration with the number of operations that would survive without it,
smallest first — and it is arithmetic, not ranking.
Below the choose threshold it stops refusing and becomes a warning instead — order_is_weak on
the delivered set. Refusing made sense while the search was deciding; handing over every candidate
is not deciding, so the disagreement reverts to being evidence about the order, which is the only
thing it was ever evidence about.
The applicability filter above runs first for both engines — otherwise the guarantee would only be true of one of them, and there is a test that says so.
Then it runs, tool or no tool. Most catalog operations have no dedicated tool:
56 of 81 are declared without one. Available operations run by name through
run_operation(operation, arguments). This is deliberate: capability count has no ceiling, but the exposed tool list has one, so the catalog
is allowed to grow faster than the tool list. Arguments are checked against the catalog before
anything executes — unknown operation (with a "did you mean", from the same ranking), missing or
misnamed argument, wrong type, path outside the workspace — and every error carries a stable
code. Execution goes through the same path as execute_plan, so an operation cannot behave one
way alone and another way inside a plan.
Both engines embed the identical document text (catalog.document_text), so a comparison
between them measures the ranking and nothing else. Three test files keep the rest under
measurement rather than under opinion: the degradation curve over our own catalog, the projection
against 800 real neighbouring operations, and a discoverability contract per entry. That is what
turns the scaling limit into a curve you can watch rather than a number someone guessed.
Determinism is the reason for building it this way rather than reaching for a hosted
embedding API: that would make tool discovery a network call whose answer can change under
you, and an agent that finds a different tool tomorrow for the same question is not
reproducible, whatever its manifest says. The one network access left is the model download
on first use, at the pinned revision; after that the vector engine is local, and an install
that never makes it keeps BM25, and the engine field of every result says which one
answered.
Making it better with your own requests, without a model that drifts
The 155 requests behind those percentages were written by two language models. They are the best set we could build without users, and they are not what users ask: a real request names the file somebody actually has and the words their field actually uses.
So MapSmith can record its own. Set MAPSMITH_DISCOVERY_LOG to a file path and each
search is written as one JSON line together with the operation that was run after it —
the query, the facets declared, which engine ranked it, every candidate delivered, and
where in that list the chosen one sat:
MAPSMITH_DISCOVERY_LOG=/data/discovery.jsonl # then work normally for a while
python benchmarks/log_to_cases.py /data/discovery.jsonlFor the part that needs eyes rather than a pipe, there is a dashboard — see below.
log_to_cases.py prints those lines as rows shaped like tests/data/discovery_queries.json
and flags the two that matter: a run the ranking did not put first (the answer was on
screen and the order was wrong) and a search nothing followed (a request the catalog did
not serve). It prints; it never writes. Which rows become test cases is a person's call.
None of this trains anything, and that is the design. A ranker that learns from what
callers pick learns from an ordering it produced: the operation shown first gets picked
more, gets learned as correct, gets ranked first harder — a confident answer nothing
contradicts, which is the exact failure this product exists to measure. The model revision
stays pinned, held there by a golden-vector test, so the same query gets the same answer
next year. What improves instead is the catalog text — a phrasing, a distinguishes that
does not distinguish — as a diff somebody can read and revert. That loop is not the weak
option: it is what took found@3 from 18% to 58% and delivery to 99%.
The log is off unless the variable is set, holds queries and operation names and nothing
else (no dataset paths, no arguments), is guarded by MAPSMITH_WORKSPACE like any other
path MapSmith writes, and never leaves the machine — nothing reads it back. Your queries
describe your work; treat the file that way, and delete it when you are done.
Formats
Format | Read | Write |
GeoParquet 1.0 / 1.1 — WKB plus | yes | yes, every path |
GeoParquet 2.0 — Parquet-native | yes, including files that carry no | yes on the SQL path: |
GeoPackage, Shapefile, FlatGeobuf, GeoJSON, … | anything pyogrio/GDAL opens | via GDAL |
GeoTIFF / COG | yes | outputs of the |
GeoParquet 2.0 moves geometry
into Parquet's own logical types and makes the geo key optional, so "a Parquet file with
geometry in it" no longer implies that key. MapSmith reads the CRS from the logical type
when it is the only place it exists — the spec default, an authority string,
projjson:<key>, or the whole PROJJSON document inline, which is what DuckDB writes.
run_sql emits both layers (geoparquet_version 'BOTH'), so one output file satisfies a
2.0-native reader and a GeoPandas 1.x one; the GeoPandas writer path stays 1.x because
GeoPandas 1.1 caps schema_version there.
One declaration is deliberately refused rather than guessed: srid:<n>. The spec defines
it as a numeric identifier and names no authority — its own example is srid:0 — so
reading it as EPSG:<n> would be inventing a coordinate system and recording it as fact.
Choosing the stack, and never swapping it in silence
MAPSMITH_STACK picks the geoprocessing stack once, at the start. The default is
opensource — GDAL, GeoPandas, DuckDB, Whitebox — and needs no licence. esri routes to
ArcPy on a machine that has ArcGIS Pro installed, through a subprocess and files, because
ArcPy lives in Pro's own interpreter. MapSmith ships no part of it and takes no licence to
look: what a session can reach is read from the metadata the installer left on disk, and
server_info reports it, so a caller learns it before planning five steps around it rather
than at the first failure.
The rule that makes the choice worth making is what happens at the edges. When the chosen stack cannot do something, MapSmith says which of three things is true — there is no such tool, this licence tier does not include it, or it would need an online service — because those lead to three different decisions and one word for all of them leads to none. Where a route exists and cannot run, the manifest names the engine that actually produced the numbers and why the preferred one did not: an engine quietly replaced by another is a record that is true and a number nobody chose.
One operation is routed today: buffer_layer. Everything else runs on the open source
stack whatever MAPSMITH_STACK says. That is the state of the wiring rather than a property
of the design, and it is written here because the alternative was a defect: a table
declaring three routes while one operation consulted the router meant requesting the stack
ran two of them elsewhere with nothing in the manifest saying so. It was fixed before this
release by shortening the table, not the sentence.
Two things this is not. It is not a comparison: MapSmith calls what you have installed, and this repository publishes no scores for anybody's engine but the ones Argleton grades in public. And it is not equivalence: two further operations were measured against both stacks on the same fixture and deliberately left unrouted, because matching geometry is not the whole story — on a dissolve the other stack drops every attribute, so a pipeline that dissolves and then reads a column would find the column on one stack and nothing on the other. That is a difference to record before it is a route to offer.
Verification, in and out
Every tool that writes a dataset also writes <output>.provenance.json beside it and
verifies its own work — CRS agreement, geometry validity, raster dimensions, count and
extent invariants — recording the results in the manifest before raising anything, so
the audit trail survives the error. A run that fails before writing, where an earlier
run's dataset still sits at the path, leaves that run's record in place and writes its own
as <output>.failed.provenance.json; the error names it.
Verification runs on the way in as well. Before an operation touches your data, MapSmith checks the failures that produce plausible junk: an input with no CRS is refused outright, because metric maths on unknown units is how a confidently wrong answer gets made; an empty input, or two layers whose extents cannot possibly overlap, comes back as a named warning with a hint — in the tool result, not only in the manifest, so the agent sees it instead of assuming success. (The join fast paths, DuckDB and SedonaDB, only ever receive inputs that already share a known CRS; they verify their output and diagnose an empty join.)
An output whose geometry is mechanically broken — typically invalidity inherited from an
invalid input — is repaired deterministically: make_valid, at most two rounds, written
to a temporary file and swapped in only once it is complete, and skipped rather than
risked where a rewrite could drop data (a multi-layer GeoPackage is refused, not
rewritten — extract_layer copies the one you mean into its own dataset, with the
container and the layers left behind named in its manifest). Every attempt lands in the manifest and in the tool result, because a
repaired output must never look like one that was right the first time. Failures that need
judgement are never "fixed": an empty result, or geometries eroded away by a wrong
distance, come back as warnings with hints for the agent to act on.
And a manifest can say which configuration produced the numbers — environment,
section 3.8 of the manifest specification, empty when there is nothing to say. What made it
concrete: a GeoTIFF and the .aux.xml file beside it can declare different
georeferencing, and GDAL prefers the sidecar by documented design, because that is how
somebody overrides georeferencing they know to be wrong. Both readings are the library
behaving exactly as written, and on one fixture the same file gives an area four times
larger and an origin a hundred kilometres away. There is nothing upstream to fix and
everything to state, so describe_dataset reports both sources when a raster has two, and
fourteen operations that read a raster's grid directly refuse instead — zonal statistics,
resampling, clipping, reclassification, band maths, reprojection, band extraction, band
statistics, locating an extreme cell, sampling at points, the elevation profile and the line
of sight — naming both readings and saying how to choose.
Describing is different from computing: a file with two georeferencings is a thing to be
told about, not a coin to flip. This is the multi-layer refusal
(#29) on a second axis — the format's
default answering a question the caller never asked.
The terrain and sampling operations do not refuse yet, and saying "any operation that computes" would be the promise-with-no-caller this release already found once: the terrain engine catches the same file by a different route, because it compares its own reading of the grid against GDAL's and stops when they differ, but its message names neither the sidecar nor the way out. Sampling a raster at points does not catch it at all. Extending the refusal to the remaining raster operations is on the roadmap below.
See results inside the chat

preview_map renders your layers on an interactive map panel inside the chat — pan,
zoom, toggle layers, and read each layer's provenance card (operation, engine, and one of
three honest states: verified ✓, verification failed, or not verifiable when no
critical check ran) right next to the geometry it explains. Field-tested on Claude
Desktop; it renders in any client that implements the official
MCP Apps extension, and on
clients without it the same call returns the preview as structured data.
The panel is self-contained — no CDN, no bundled libraries, no telemetry — with one outbound request named here rather than buried: the OpenStreetMap background tiles, which reveal the map view you are looking at (never your data) and which the panel drops to a plain backdrop when the host blocks them. The preview is deliberately lossy (simplified geometry, capped feature counts): the dataset of record stays on disk with its manifest.
Plans: reject malformed analyses before they run
In GISAgentBench — 349 practitioner-sourced tasks over 128 GIS APIs — the best frontier agent completes 32.7% of tasks under strict scoring, and planning defects dominate the failures: missing operations in 28.3% of failed runs and wrong operation order in 18.4% (multi-label, so up to ~47% involve a planning mistake), against 7.8% for parameter errors. MapSmith attacks this where it is cheapest: the agent submits a typed plan, and static validation rejects unknown operations (with suggestions), missing arguments, forward references, absent input files and CRS-unsuitable steps before anything executes — with machine-actionable error codes the agent can repair.
What it does not do, said here because the heading used to claim it did. Validation is structural. It reads the shape of the plan, not its meaning, so a plan that is well formed and answers the wrong question passes. Demonstrated on this repository's own worked example on 2026-09-14: asked to clip the parcels before buffering the river, MapSmith did exactly that. The intermediate is a 1.5 km circle centred on one parcel corner, it inherits that parcel's name so downstream it looks like a parcel, every step reported verified, and on this fixture the final table comes out identical to the correct one — the defect is silent in the output. The heading said "reject wrong analyses" until that run; it says malformed now, because that is what is true. Catching the other kind is an open problem, and it is the reason the correctness suite exists — though this particular wrong-order case is outside its current probes, so today it does not catch this one either.
{
"goal": "buildings within 300 m of rivers",
"steps": [
{"id": "buf", "operation": "buffer_layer",
"arguments": {"input_path": "rivers.gpkg", "distance_meters": 300,
"output_path": "rivers_300m.parquet"}},
{"id": "cut", "operation": "clip_layer",
"arguments": {"input_path": "buildings.parquet", "mask_path": "$buf",
"output_path": "at_risk.parquet"}}
]
}"$buf" consumes the output of step buf; references may only point backwards, so plans
are acyclic by construction. validate_plan also simulates the CRS of every intermediate
dataset from the real input files. execute_plan then runs the chain with per-step
provenance plus a plan-level manifest (<output>.plan.json) fingerprinting the exact plan
that produced the result.
Confinement
UNC hosts and NTFS alternate data streams are refused in every path argument of every
tool call, before anything touches the filesystem (on Windows even an existence check on a
UNC path talks to an attacker-chosen host). Remote and virtual forms — GDAL /vsi*,
https:// COGs — are refused by default since 0.2.2 and need MAPSMITH_ALLOW_REMOTE=1;
a workspace refuses them whatever that setting says (details below). Validated plans are
stricter by design and reject every non-local form, opt-in or not.
Set MAPSMITH_WORKSPACE=/data to confine the server to one directory:
every path argument of every tool must resolve inside the workspace (checked at the MCP boundary, and again by plan validation with stable error codes);
the coordinate system an input declares is read from its bytes before GDAL opens it, because GDAL would open any grid file it names, for every format MapSmith accepts, and a format it does not recognise is refused -- netCDF, HDF, JPEG 2000, ERDAS IMG and GRIB rasters among them, until MapSmith reads their coordinate system itself (since 0.10.1). JSON, text and individual ZIP members over 32 MiB are refused before a whole-file read;
the
run_sqlDuckDB connection is sandboxed in the engine itself, because SQL text is out of reach of a textual path check: filesystem whitelisted to the workspace (allowed_directories+ external access off, which also covers GDAL-backedST_Read), memory and temp disk capped (MAPSMITH_DUCKDB_MEMORY, default 4GB;MAPSMITH_DUCKDB_TEMP_LIMIT, default 8GB), configuration locked. DuckDB's spatial extension carries a PROJ of its own that this filesystem never sees, so the statement is also read with DuckDB's parser before it runs: onlySELECT, and every coordinate system (ST_Transform,ST_SetCRS) and dataset path (ST_Read,read_parquet) as a literal that passes the same checks as a tool argument, and through GDAL only GeoPackage, OGR SQLite and SpatiaLite with MapSmith-read SRS catalogues, shapefile and GeoParquet, whose coordinate system MapSmith reads first (since 0.10.0; SECURITY.md has the hole it closes).
MAPSMITH_DISCOVERY_LOG is one of three paths MapSmith writes to that no tool argument
names. The other two are written whether or not anyone asks — DuckDB's extension directory
and the scratch directory engines use inside the workspace — and SECURITY.md names them
in its Workspace containment promise; this one is off unless you set it, and SECURITY.md
covers it under What MapSmith records. It goes through the same check as a tool
argument: outside the workspace it is refused, and the refusal disables the log and says
so on stderr rather than failing the search that triggered it.
Without a workspace, file access is deliberately unconfined — fine for a local stdio
server on your own files — and plan validation flags run_sql steps with a
SQL_NOT_SANDBOXED warning. Code execution is closed in both modes, and since 0.4.0 the
layer that closes it is the right one: INSTALL and LOAD in a statement are refused
outright, because an INSTALL is an HTTPS fetch of a native binary run in this process on
SQL a model wrote. Until 0.4.0 only the implicit forms were off, and an audit installed
DuckDB's aws extension and read this machine's real cloud credentials back through a tool
result — the whole story is in SECURITY.md, including why none of the four
existing layers saw it. Extensions already loaded keep working, spatial included; to
acquire others, name them where the agent cannot reach:
MAPSMITH_ALLOW_EXTENSIONS=postgres,azure in the environment of the process that starts
the server. Behind that, community extensions stay off (shellfs turns a filename into a
shell command), unsigned extensions are refused, DuckDB's HTTP and S3 filesystems are
disabled, and the configuration is locked.
The network is closed too, unless you open it. Remote and virtual forms — GDAL
/vsi*, https:// COGs — are refused by default in path arguments and inside run_sql
text, because the path is written by the model rather than by you: a third-party dataset
carrying "the updated layer lives at https://evil.tld/x.gpkg" was otherwise enough to have
GDAL parse attacker-chosen bytes in-process. Set MAPSMITH_ALLOW_REMOTE=1 to allow them
— cloud-native data is a real use case and the capability is gated, not removed. A workspace
refuses them regardless, since containment and "fetch whatever URL the model names" cannot
both be true. The test suite asserts every branch by counting requests at a loopback server
(tests/test_duckdb_sandbox.py). The full threat model —
and what is explicitly not covered — is in SECURITY.md.
Fine print, because it changes how you deploy this: the path jail assumes a single trusted
writer of the workspace filesystem (paths are resolved at check time, so a symlink swap by
another local process is out of scope); the DuckDB spatial extension is fetched once per
environment by MapSmith itself, through the Python API rather than by any statement, so on
air-gapped machines pre-install it (python -c "import duckdb; duckdb.connect().install_extension('spatial')") before locking the network down; and the
HTTP transport has no authentication in this release, so keep it on loopback or a trusted
network. For real isolation, run the container and mount only the data you want it to see.
The dashboard
Everything this project knows about itself is computed somewhere and most of it is printed
once and lost, which is how a number ages into a claim. benchmarks/dashboard.py gathers it
into one self-contained HTML file — no CDN, no fonts, no analytics, works with the network
off:
python benchmarks/dashboard.py --log /data/discovery.jsonl --argleton ../argletonOperations — every entry, and whether a caller actually reaches it. Asked twice: with words alone, and with the facets a caller knows. Three outcomes are kept apart — a rank, an answer that did not contain the entry, and a search that declined because the two rankers shared nothing. Collapsing the third into the second is the first thing this page got wrong about itself, and it drew ten working operations as broken.
Search quality — the facet ablation for both rankers, and the degradation curve as the catalog grows, which is the measurement the embedding engine became a dependency for.
Traps — Argleton's families and what each engine does with them, MapSmith included and not flattered. With
--argleton <path>it reads a checkout and shows the per-family detail; without one it falls back to the vendored citation and says so.Answer the open questions — the requests where the two model labellers named different operations, and the cases the discovery log recorded, answered by clicking. The percentages recompute against your answers as you give them, and they come back out as JSON. This is the open question the published figures rest on: two labellers agreeing 75% of the time is the ceiling of a task with no single right answer, and the only way past it is somebody who has done the job.
Trend — each generation appends a row beside the page, so the numbers are a series rather than a snapshot. Identical consecutive rows are dropped: rebuilding five times must not manufacture a trend.
It is a snapshot — a new operation or a new trap appears when it is generated again — and regenerating keeps every answer already given, because answers are stored against the text of each question rather than its position.
We measured whether this actually helps
Claims about agent performance are cheap, so docs/benchmarks.md reports an A/B on GABench — 57 executable GIS tasks over a 133-tool server, scored by its deterministic evaluator — where the only variable is whether the agent's typed plan is validated before the solver runs. GABench is by Bo Yu et al. (arXiv:2604.13888); the counts here are of the checkout we ran and differ from the paper's, which docs/benchmarks.md states side by side.
The honest headline is a null result, on a frontier model and on a small one, and the interesting part is why:
Arm A (no gate) | Arm B (gate) | |
Sonnet 5 — TAO / PEA | 0.824 / 0.430 | 0.781 / 0.425 |
Haiku 4.5 — TAO / PEA | 0.660 / 0.320 | 0.714 / 0.366 |
Haiku looks like a clean win until you notice the gate only fired on 4 of 57 plans, and that the 53 tasks it never touched moved by just as much: the aggregate delta is run-to-run variance, and measuring that noise floor (2–5 points per metric on a single repetition) is the reusable result. What survives is narrower — on the plans it did repair, tool selection improved by +0.19 TAO — and it points at where the failures actually are: PEA around 0.4 in every arm, i.e. wrong parameters and missing outputs at execution time, which is why MapSmith enforces its plans at the execution boundary and verifies inputs and outputs at runtime rather than advising an agent that improvises.
Three further arms then measured the configuration MapSmith actually ships — the plan enforced, no improvisation between validation and execution — over 375 runs, and the result cuts both ways: enforcing reproduces its own score 3–18× more tightly than an improvising solver, and it does not beat it on accuracy (parity on tool selection, measurably worse on ordering). One of those arms also refuted a conclusion this page had published two arms earlier; the correction is kept in place rather than edited away.
The harness is in benchmarks/gabench-ab/, including
the split_analysis.py that took our own win apart and the
rep_analysis.py that bars every delta against a measured noise floor.
What the suite found in MapSmith
This section is the one a reader should be most suspicious of, so it is a heading rather than a line inside the roadmap, where it used to be. Everything in it is somebody grading us, and the grader is ours.
Argleton is a correctness suite for a result that is wrong and reported as successful, with the right answer known by construction. It is ours, and it lives in its own organisation under Apache-2.0 rather than in this repository, because an evaluation that lives inside the thing it evaluates is easy to dismiss in one line. Closed-form truth, no model in the evaluator, fixtures rebuilt rather than vendored.
The score. On the current run — all thirty families — MapSmith answers every
trap correctly: 0.00 silent errors over 33 traps, nothing skipped, against 0.8788
for a careless composition of the same libraries. The runs from 2026-08-30 to 2026-09-15 said
0.00 as well and were wrong on one trap, a point-registered DEM: its truth was derived from a
premise about GDAL that GDAL documents the other way, MapSmith had been built to that premise,
and both corrected half a cell that was already corrected. Found on 2026-09-25 by a Copernicus
DEM whose samples fall on whole arc-seconds by its own documentation; the
erratum rescores every run it touched, and the fix shipped in 0.7.0; every run since measures it. One of the thirty-three is not an answer
at all but a refusal: a raster and the sidecar beside it declare different georeferencing,
both readings are GDAL behaving as documented, and the right move is to stop and say so
rather than pick one. That family arrived on 2026-08-31, the morning after the list of
twenty-seven was closed, and it is the reason a manifest can now name its environment
(see the changelog).
The twenty-ninth family arrived on 2026-09-02 and is the one that reads oddly until you see it: a survey plot whose easement ring is wound the same way as its outline. A shapefile carries no nesting, so which ring is a hole is decided by the winding and by nothing else — the easement comes back as a second shell and its area is added, flattering the owner by 6.9% with a correct bounding box, a correct CRS and no warning. MapSmith repairs it and records the repair, which is the only reason the number is right.
The thirtieth family arrived on 2026-10-29: a CRS declares its own datum shift, with accuracy unknown rather than a ballpark transformation. Treating that unknown as a reason to refuse would discard an explicit Helmert transformation and produce no answer where a shifted coordinate is available.
The finding that matters more than the score, and it is about us: on the first run that 0 and MapSmith's verification had nothing to do with each other — seven checks passed on that trap and not one of them looks at whether the number is right. A provenance manifest records what was done; it does not certify that it was right, and this README used to imply otherwise by promising a run "with verification disabled". There is no such switch and we are not adding one.
The run also separates the passes MapSmith earned from the ones it did not. The
mismatched-CRS join and the feet-as-metres unit are its own discipline; the Web Mercator
pass comes from a default (ground area is geodesic unless you ask for the plane) rather
than from care; the TIFF-predictor pass is still rasterio's. The datum-ballpark pass is
the newest and the least flattering: MapSmith failed that trap on 2026-08-26 — 74 m
out, with a manifest recording a successful reprojection — and the pass is the fix, not
the original behaviour. The run where it failed is still published.
Seven defects have come back from it, which is the return we wanted from putting the suite outside:
the
datum-ballparkfailure above — 74 m out with a manifest recording a successful reprojection, the most serious of them because nothing in the output looked wrong;a multi-layer container resolved silently to its default layer, answering 4 features where the truth was 31 (#29, filed before the trap was published —
extract_layeris the way out of that refusal now);a south-up DEM whose georeferencing was dropped on read, so a 5.71° slope came back as 45° with all five verification checks green — the coordinate system had survived and only the geotransform had not;
totals over a layer holding lines and polygons together, where a treatment plant's fence line was added to 2 km of pipe in silence and every individual row stayed right;
two probes answered
unsupportedbecause nothing could say where a raster's lowest cell was;three probes that came back
unsupportedbecause MapSmith had no area operation at all —measure_areaexists because a trap said so, and it carries the first check here that asks whether the number is right rather than whether the operation ran;and the seventh, which the suite sent back by being wrong itself. Grid registration was the family MapSmith could not attempt when it was published, and the fix for it was built to the trap's premise: that GDAL leaves a point-registered file's geotransform alone, so the reader must move every position half a cell. GDAL documents the opposite (RFC 33) — it has already moved it — and MapSmith put every sample of such a file half a cell north-west, the Copernicus DEM included, while the trap scored that as a pass. A Copernicus DEM, whose samples fall on whole arc-seconds by its own documentation, found both on 2026-09-25 (erratum). Positions now come from GDAL's geotransform whatever the tag says, and the terrain engine, which really does misread the tag, gets an area copy instead of a refusal.
Example gallery
Five executable walkthroughs in examples/: verified buffer+clip with
provenance manifests, terrain and hydrology on a real 520×520 USGS DEM of Mount St.
Helens, a deliberately wrong plan rejected before execution and then repaired,
Copernicus data on Monte Baldo — a DEM in degrees refused for slope and then reprojected,
and a Sentinel-2 NDVI whose reflectance offset the catalogue declares and the pixels already
contain — and an agentic pilot that turns a question into a
validated plan, a QGIS execution and a verified 1,300 m² answer. The
terrain notebook also shows what happens when reality bites: that DEM is stored with the
standard TIFF predictor, which Whitebox Workflows 2.x does not undo when reading
(upstream report), so MapSmith
detects it, converts the input first, and discloses the workaround in the manifest.
Architecture
AI agent (Claude / ChatGPT / Copilot / your app)
│ MCP (stdio local · Streamable HTTP remote)
▼
┌─────────────────────────────────────────────┐
│ MapSmith server │
│ · semantic tools + operation catalog │
│ · parameter validation, CRS discipline │
│ · provenance recorder (lineage manifests) │
├─────────────────────────────────────────────┤
│ Engines │
│ · vector: GeoPandas/Shapely (built-in) │
│ · SQL/analytics: DuckDB Spatial (built-in) │
│ · heavy joins: SedonaDB ([sedona] extra) │
│ · zonal stats: exactextract ([raster]) │
│ · terrain/hydro: Whitebox NG ([whitebox]) │
│ · QGIS native:buffer via subprocess │
│ (GPL-isolated; GRASS sidecar: roadmap) │
└─────────────────────────────────────────────┘When not to use MapSmith
You need an authenticated remote server today. The Streamable HTTP transport has no authentication in this release: anyone who can reach the endpoint can run every tool against everything the process can see. Loopback or a trusted network only (SECURITY.md).
You want a sandbox for arbitrary agent code. MapSmith confines paths and the SQL engine; there is no code-execution tool yet, and a path jail is not a container.
You need cartography. No styling, no layouts, no print composer. Outputs are datasets, plus a lossy read-only preview panel — not maps you publish.
Your data lives in a database. MapSmith reads and writes files (GeoParquet, GeoPackage, anything GDAL opens). There is no PostGIS engine and no database catalog — the
[postgres]extra is for the optional job ledger, not for data.Your data lives in object storage. Since 0.2.2 remote and virtual paths are refused unless you set
MAPSMITH_ALLOW_REMOTE=1, and refused whatever that setting says under a workspace — which is what the container runs with by default. DuckDB's own HTTP and S3 filesystems stay off in every mode, soread_parquet('s3://…')does not work even with the opt-in: fetch the data down first, or run unconfined with remote reads on.You want the full breadth of a desktop GIS. 81 operations, and a catalogue that tells the agent what does not exist yet rather than letting it improvise. The ~900 QGIS Processing algorithms are on the roadmap, not in the box.
You expect plan validation to make a weak model strong. Our own A/B says advisory validation upstream of an improvising solver does approximately nothing at aggregate level — and the enforced configuration MapSmith ships, measured afterwards, did not beat it on accuracy either. What enforcement buys is reproducibility (the numbers).
You want us to debug your local geospatial toolchain. Docker, or
uvxwhere the wheels work, are the only supported paths; a hand-built native GDAL stack is not, on purpose.
Roadmap
Zonal statistics (exactextract, exact fractional coverage)
Whitebox Next Gen adapter: hillshade, flow accumulation, watershed (in-memory, open tier)
Typed analysis plans: static validation against the operation registry + simulated CRS flow before execution
Runtime verification: input preconditions, warnings with hints in the tool result, bounded deterministic repair recorded in the manifest
MCP Apps in-chat map panel with provenance cards (self-contained, works under the default host sandbox)
GeoParquet 2.0: read Parquet-native geometry types (including files with no
geokey), write both layers from the SQL path — the GeoPandas writer path follows when GeoPandas lifts itsschema_versioncap and 2.0.0 stops being a release candidate
Next, in the order we intend to do it. The linked items carry a written spec — a roadmap line without one is a wish, so the rest get theirs before work starts on them:
A suite for plausible results reported as successful when they are wrong — with the right answer known by construction. It exists, it is not here, and it is not ours to grade: Argleton lives in its own organisation under Apache-2.0. What it measures in MapSmith, including the seven defects it has sent back, has a section of its own. #25 is closed against Argleton rather than left open here.
Agent-loop repair: hand verification failures back to the agent as structured, actionable errors, with a bounded retry budget recorded in the manifest. Our own measurements say the runtime error message is the information channel that works
Tool contracts that carry their own rules: argument constraints enforced and stated, and errors that name the rule rather than only the violation. The one intervention in our benchmark work that moved a metric past its noise floor
A project brief for the requests that are a chain, not a call. A third of real requests are not one operation: on the fifty hardest requests in our own set — the ones two model labellers disagreed about, answered by hand on 2026-09-01 — sixteen have "this is a sequence" as the answer or as a defensible second answer. To those, handing back a set of candidates is the wrong shape: none of the candidates does what was asked. So the answer becomes a brief: what has to happen, in what order, which engines this machine has, and which decisions the caller has to make before anything runs — the extensive-or-intensive choice in an areal interpolation, the buffer width that lives in a regulation rather than in the request, the base length of a gradient. Today those surface one at a time, as refusals.
Two constraints are part of the plan rather than details of it. The brief is a rendering of an already-validated plan, never a preamble to one: build the plan, validate it, then narrate the validated plan — and every sentence in it has to trace back to a declared field, so a claim that traces back to nothing is a claim somebody invented, which is checkable by machine. And MapSmith installs nothing: the brief names what is missing and prints the command, and a person runs it. The reason is dated — see SECURITY.md on why
INSTALLin a SQL statement is refused in both modes since 0.4.0. A server that installs on authorisation is the same shape with a consent dialog in front, and the consent comes from somebody who has just read persuasive prose written by the process asking for it.Depends on the item above: the brief reads the rules the tools declare, so there is less to derive from without them.
Satellite embeddings as a first-class input: per-zone embedding vectors (multiband zonal statistics) and similarity rasters against a reference location, over the open AlphaEarth annual dataset (CC-BY 4.0 COGs). Deterministic arithmetic on a raster — no model inference in MapSmith — with the tile, year and reference vector recorded in the manifest
Authenticated remote mode (OAuth on the existing Streamable HTTP transport) and long-job progress via MCP Tasks. This is the item that closes the one limitation SECURITY.md declares outright: the HTTP transport has no authentication today
Slope and aspect (Whitebox, closed-form tested; geographic-CRS DEMs refused)
Stream network extraction (Whitebox, from a flow-accumulation grid; the threshold and its unit recorded in the manifest)
More terrain & hydrology: curvature (six kinds, the kind required because profile and plan answer opposite questions), flow direction (d8/rho8/dinf/fd8, with the direction-code table written into the manifest — the engine's own manual documents its default table backwards, so a name would not have been enough), Euclidean distance and IDW interpolation
The ambiguous-georeferencing refusal on every raster operation rather than the fourteen that read the grid directly. The three sampling operations were added on 2026-09-02 —
sample_raster_at_pointswas the worst gap in the product, since the same DEM answered 10.0 and 30.0 with no sidecar and 2.0 and 6.0 with a 40 m one, five times out with no warning. What is left is the terrain engine, which stops on the same file for a different reason — its own reading of the grid disagrees with GDAL's — and whose message names neither the sidecar nor the way outMap panel: MapLibre vector rendering, and an export of the panel as a self-contained HTML file you host yourself (raster OSM tiles already ship). No hosted viewer — MapSmith runs on your machine and we would rather not own your maps
Not planned, and closed as such on 2026-09-01: a sandboxed code-execution tool for the long tail. A typed plan is the same efficiency win in a shape that can be refused for a stated reason before anything runs; a model-authored script keeps the arithmetic in the engines but moves the composition — engine, order, units, CRS — into text nobody validated, and then emits a manifest that is true about the library and silent about the part that decided
QGIS Processing sidecar (subprocess-isolated), with
native:bufferverified end to end and other algorithms added only through the same allowlist, parameter mapping and provenance
License and project
MapSmith server and engines: AGPL-3.0-or-later (see LICENSE)
Client SDK and tool-schema definitions (future
sdk/): Apache-2.0
You can self-host MapSmith freely, forever. If you modify it and offer it as a service, the AGPL asks you to share your changes — or talk to us about a commercial license.
Nothing here has been funded so far. funding.json lists, in the
FLOSS/fund format, the two pieces of work money was asked for:
a public suite of geospatial traps with hand-computable answers, and the provenance
manifest as a specification other tools can implement. Both were built anyway, unfunded,
and are archived with DOIs; the entries stay in the file marked inactive, so the estimate
can be read against what came of it.
Release notes are in CHANGELOG.md, how to contribute in CONTRIBUTING.md (AI assistants start with AGENTS.md), how to report a vulnerability in SECURITY.md. "MapSmith" is a trademark of the MapSmith project — see TRADEMARKS.md. Updates: LinkedIn · X · Bluesky.
Available Tools
29 toolsaspectAIdempotent
Aspect from a DEM: downslope azimuth in degrees, 0 = north. GeoTIFF in/out.
FLAT CELLS ARE -1, not nodata — mask them before averaging aspect over an area, or the average is plausibly wrong. DEMs in a geographic CRS are refused (see slope): reproject to a projected CRS first. Requires the [whitebox] extra.
Inputs: dem_path (required), output_path (required), z_factor. Returns: object
brief_planARead-onlyIdempotent
Render a project brief from a plan that has ALREADY passed validation.
Every statement carries its source: operation declaration, argument rule, validation finding, engine availability or install requirement. Invalid plans return only the validation report and no project text. Installation commands are printed for the person to run; MapSmith never installs an engine.
Inputs: plan (required). Returns: object
buffer_layerAIdempotent
Buffer all features by a distance in meters.
Geographic-CRS inputs are reprojected to an estimated UTM zone for the metric
operation and back; the decision is recorded in the provenance manifest.
A warnings key in the result flags a suspicious-but-valid outcome with a
hint — e.g. a negative distance that eroded every geometry away.
Inputs: input_path (required), output_path (required), distance_meters (required). Returns: object
centroid_layerAIdempotent
One point per feature: the geometric centroid, computed in a metric CRS.
Geographic-CRS inputs are measured in an estimated UTM zone (decision recorded in the manifest) and returned in the input CRS — a planar centroid of degree coordinates lands in the wrong place, quietly. The output is verified: same feature count, Point geometry, input CRS. Note the manifest's caveat: the centroid of a concave or multi-part feature can fall outside it. Inputs without a CRS are refused.
Inputs: input_path (required), output_path (required). Returns: object
clip_layerAIdempotent
Clip a layer to the area of a mask layer. CRS are aligned automatically.
A warnings key in the result means the analysis ran but something is worth
your attention (typically an empty result, or inputs whose extents do not
overlap); each entry carries a hint. Inputs without a CRS are refused.
Inputs: mask_path (required), input_path (required), output_path (required). Returns: object
convert_formatAIdempotent
Convert a vector dataset between formats; the target is chosen by the output extension (.parquet, .gpkg, .geojson).
The output is re-read and verified: same feature count, same CRS. Two conversions are refused with the reason: shapefile (field names truncated to 10 characters, silently) and GeoJSON for non-WGS84 layers (RFC 7946 is WGS84 by definition — reproject first). Invalid geometry carried through is repaired deterministically and reported in a 'repairs' key.
Inputs: input_path (required), output_path (required). Returns: object
describe_datasetARead-onlyIdempotent
Inspect a dataset, vector or raster, before analysing it.
Vector: CRS, geometry types, schema, extent, feature count. A MULTI-LAYER container (e.g. a GeoPackage holding several layers) is described per layer — name, feature count, geometry type, CRS — because operations refuse containers with no chosen layer: extract the layer you mean first (run_sql: SELECT * FROM ST_Read(path, layer='name') with an output_path). Raster (.tif): CRS, grid size, resolution, bands with dtype, nodata and masked statistics (nodata cells counted separately). Call this first on any dataset you have not inspected yet — most silent GIS errors start with wrong assumptions about CRS, units, nodata or which layer you are on. Raster inspection requires the [raster] extra.
Inputs: path (required). Returns: object
dissolve_layerAIdempotent
Merge features into one geometry per value of by (or one feature in all).
aggfunc — first (default), last, sum, mean, median, min, max or count — is
applied to the other columns and RECORDED in the manifest: a sum reported
where a mean was meant is a plausible wrong number nobody can see. Features
with a null by key are dropped by the grouping and the manifest counts
them. The output feature count is verified against the number of distinct
keys, so a wrong grouping fails loudly instead of shipping.
Inputs: input_path (required), output_path (required), by, aggfunc. Returns: object
execute_planAIdempotent
Validate, then execute a geoprocessing plan step by step.
The plan is re-validated first (an invalid plan runs nothing). Steps run in order; "$step_id" references resolve to the outputs of earlier steps. Every step writes its own provenance manifest, and a plan-level manifest (.plan.json, with the plan sha256 and per-step outcomes) ties them together. Execution stops at the first failing step; outputs already produced stay on disk with their manifests. Same plan format as validate_plan — validate first, then execute.
A step_warnings key in the response means the plan ran but some step
produced a suspicious result (an empty output, non-overlapping inputs):
read it before treating a completed plan as a correct one.
Verification failures include structured checks and a three-attempt repair budget. To retry, change the plan yourself, choose NEW output paths and pass the previous .plan.json as repair.previous_manifest. MapSmith never chooses or runs the repair for you; every attempt is linked in the manifests and a fourth is refused.
Inputs: plan (required), repair. Returns: object
explode_layerAIdempotent
Split multi-part geometries into one feature per part (attributes copied).
The output feature count is verified against the number of parts counted before the engine ran, so a lost part fails loudly instead of shipping. Inputs without a CRS are refused.
Inputs: input_path (required), output_path (required). Returns: object
flow_accumulationAIdempotent
D8 flow accumulation from a DEM (GeoTIFF in/out). Depressions are filled first.
out_type: 'cells' (upslope cell count, includes the cell itself) or 'sca' (specific catchment area). log_transform=True for visualization-friendly values. Requires the [whitebox] extra.
Inputs: dem_path (required), output_path (required), out_type, log_transform. Returns: object
get_provenanceARead-onlyIdempotent
Return the manifest of the ONE operation that wrote this dataset.
Its inputs with their digests, the exact parameters, the CRS decisions with
reasons, engine and version, and every deterministic check with its result.
One step, in full. For the whole chain that led here — the operations
behind the inputs, and the ones behind those — the catalogue operation
get_lineage walks it and returns each step summarised instead of one step
complete; run it through run_operation. It is not a tool of its own on
purpose: two exposed tools answering the same question about the same file
is the overlap that makes an agent pick wrong.
Inputs: output_path (required). Returns: object
hillshadeAIdempotent
Shaded relief from a DEM: GeoTIFF in, GeoTIFF out (values scaled 0-32767).
azimuth = sun direction in degrees (default 315, NW); altitude = sun angle above the horizon (default 30). DEMs without a CRS are rejected. Requires the [whitebox] extra.
Inputs: dem_path (required), output_path (required), azimuth, altitude, +1 more. Returns: object
list_operationsARead-onlyIdempotent
Find the operation you need. Say what you have and what you want — it matters more than the words you search with.
Ranking alone does not scale, and this is measured rather than assumed on independently written requests. Declaring what you already know makes ranking less important: when few enough operations survive, the whole set is handed over rather than a ranked shortlist. Delivery is not an accuracy figure: it depends on the surviving set staying small enough to return whole.
So fill these in whenever you know them, and you usually do:
input_kind — what you are holding: 'vector' (points, lines, polygons), 'raster' (a grid, a GeoTIFF), 'dataset' (either), 'plan', or 'none'.
produces — what you want back: 'dataset:vector', 'dataset:raster', 'answer' (a number, nothing written), 'description' (what something IS, rather than a computation over it), 'plan_result'.
category — the family, when you know it: vector, raster, terrain, hydrology, inspection, sql, network, planning, provenance, visualization, bridge. Unlike the others this one only ORDERS the results — a wrong guess about our families costs you positions, never the answer, so guessing is safe.
projected — pass False if your data is in a geographic CRS (degrees), and every operation that would refuse it disappears from the results.
dataset_inputs — how many datasets you are holding for this step: 1 if you have one layer, 2 if the operation combines two. This is the facet that makes a large catalog usable: on the current one it takes the surviving set from a median of 34 to 9, because "clip these parcels with that boundary" and "simplify these parcels" are different questions and you already know which one you have.
geometry — for a vector layer, what it holds: 'point', 'line' or 'polygon' (describe_dataset tells you), or a list for a mixed layer. Operations that cannot take that geometry drop out; operations that take any geometry always stay, so declaring it never hides one that fits. With two layers, list the geometry of both (points and polygons to count points in polygons). Unsure? Leave it out.
query is then plain words for what you are trying to do, and it breaks the
tie inside what is left. Describe the PROBLEM rather than the operation:
"the coastline has too many vertices and the browser dies" works as well as
the name of the tool, and better when you do not know the name.
If the answer comes back as a single entry with status: "choose", that is
the normal case and it is asking you to pick. It carries every operation
that survived, in relevance order, each with the sentence saying what it is
NOT for. The order is a hint and nothing else: our ranking puts the right
operation in the top three 75% of the time, while a model reading the same
candidates and choosing gets its first pick right 69% — and 75% is where the two
model labellers who wrote the ground truth agree with EACH OTHER, so there is
often no single right answer to rank toward. You have context no ranking has: which
file is open, what ran a minute ago, what the person actually asked for. Use
it. And if two candidates would both be defensible, ask them — that is a
better move than picking one silently, and it is what a GIS analyst would do.
A order_is_weak field means the two rankers shared nothing in their top
three, which usually means the request does not match this catalog well: read
the candidates instead of trusting the order, and say so if none of them fits.
If the answer comes back with status: "unsure", the two ranking engines
agreed on nothing and the set was too large to hand over — usually the
request was not understood rather than impossible. It carries both engines'
guesses and a question; answering the question with the facets above is the
fastest way through.
If the answer comes back with status: "none_apply", nothing you declared
can be true at once — no ranking ran. It lists each declaration and how many
operations would come back without it, smallest first, so the one that is
excluding everything is the first line. The common case is produces: several
operations compute the number you want and write it into a column instead of
returning it, so they declare dataset:vector. If nothing in relax helps,
MapSmith probably does not do this — say so rather than running a neighbour.
detail=True adds parameters and worked example calls: use it on the exact
operation name before calling an unfamiliar tool. An empty query lists
everything that survives the facets, planned operations included.
engine selects the ranker and every result says which one ran: 'auto' (the
default) prefers embeddings and falls back to BM25 where the model cannot
load; 'lexical' is BM25 alone, deterministic and network-free; 'vector'
forces embeddings. The default changed on measurement, not preference, and
the facets above matter far more than this choice.
Inputs: limit, query, detail, engine, +6 more. Returns: result
measure_areaAIdempotent
Area per feature in SQUARE METRES, written to a named column, with the total in the result.
method='geodesic' (default) measures ground area on the ellipsoid the layer's CRS names: no map plane, so no projection distortion. method= 'planar' measures in the layer's own CRS and converts with its declared linear unit — a layer in US survey feet is not assumed to be in metres — and is refused on a geographic CRS, where an area would be in square degrees.
Two things this tool does that a bare area call cannot: invalid geometry is
repaired BEFORE measuring (the planar area of a self-intersecting ring is
the signed shoelace, a number matching no region, returned without
complaint) and every repair is recorded; and a planar measurement is
compared against the ground area, so a plane that is not equal-area here
comes back with a warnings entry carrying the ratio — Web Mercator at 42°
reports 1.80× the land it covers.
Inputs: input_path (required), output_path (required), method, area_column. Returns: object
merge_layersAIdempotent
Append two or more layers into one (schema union, attributes aligned by name).
Layers are reprojected to the FIRST layer's CRS when they differ; the decision is recorded in the provenance manifest. Columns present in only some inputs are null-filled in the others and the manifest names them — data that looks measured and is actually absent is a silent error. The output feature count is verified against the sum of the input counts. Inputs without a CRS are refused. This is an append, not a geometric union: use dissolve_layer to merge geometries afterwards.
Inputs: input_paths (required), output_path (required). Returns: object
nearest_joinAIdempotent
Attach each feature's nearest neighbour from another layer, with the distance IN METERS in a named column.
Geographic-CRS inputs are measured in an estimated UTM zone (decision
recorded in the manifest) and returned in the input CRS — a nearest
distance in degrees is the classic silent error of this operation, and it
cannot happen here. max_distance_meters drops pairs farther than that; an
emptied result comes back with a warnings entry, never silently.
Inputs: left_path (required), right_path (required), output_path (required), distance_column, +1 more. Returns: object
overlay_layersAIdempotent
Set-theoretic overlay of two layers: intersection (default), union, identity, symmetric_difference or difference.
The overlay layer is reprojected to the input CRS when they differ; the
decision is recorded in the provenance manifest. Overlay pieces of lower
dimension than the inputs (shared edges, corner contacts) are dropped, and
the manifest says so. Inputs without a CRS are refused; an empty result
comes back with a warnings entry, never as a silent success.
Inputs: input_path (required), output_path (required), overlay_path (required), how. Returns: object
preview_mapARead-onlyIdempotent
Show datasets on the interactive in-chat map panel (MCP Apps).
Pass the paths of one or more MapSmith outputs or source datasets (vector or GeoTIFF). Layers are previewed in EPSG:4326 with simplified geometry and capped feature counts sized to fit client limits; each layer card shows its provenance summary and verification status. Read-only: the datasets of record stay on disk. On clients without MCP Apps support the same payload is returned as structured data.
Inputs: paths (required), max_features. Returns: object
reproject_layerAIdempotent
Reproject a layer to a target CRS, e.g. 'EPSG:32632' or a WKT string.
Inputs without a CRS are refused. Geometry passes through unchanged, so an
invalid input yields an invalid output: mechanically broken geometry is
repaired deterministically and reported in a repairs key — read it, the
geometry type may have changed.
Inputs: input_path (required), target_crs (required), output_path (required). Returns: object
run_operationAIdempotent
Run ANY catalog operation by name, including the ones with no tool of their own — which is most of them, and increasingly so.
The tools above are the handful an agent reaches for constantly. The catalog holds every operation MapSmith can perform, and it grows faster than the tool list on purpose: tool-selection accuracy degrades past a few dozen exposed tools, while capability count has no such ceiling. Discover with list_operations (use detail=true to get parameters and worked examples), then call it here.
Arguments are validated against the catalog BEFORE anything runs — unknown operation, missing or misnamed argument, wrong type, path outside the workspace — and the errors come back with stable codes, so a failed call tells the planner what to fix instead of what went wrong. Execution goes through the same path as execute_plan, so an operation cannot behave one way here and another way in a plan.
Inputs: arguments (required), operation (required). Returns: object
run_sqlADestructive
Run spatial SQL (DuckDB dialect, ST_* functions, read_parquet/ST_Read for files).
Without output_path: returns up to 50 preview rows. With output_path (.parquet): materializes the full result as GeoParquet with a provenance manifest. input_paths declares every dataset read; when omitted, the engine derives them from the query. Under MAPSMITH_WORKSPACE only SELECT runs, and coordinate systems (ST_Transform, ST_SetCRS) and dataset paths (ST_Read, read_parquet) must be string literals, so they can be checked before anything opens them; ST_Read takes there the formats whose coordinate system MapSmith reads first (GeoPackage, OGR SQLite and SpatiaLite with MapSmith-read SRS catalogues, shapefile, GeoParquet, GeoJSON, GML, FlatGeobuf, GeoTIFF, CSV...).
Inputs: query (required), input_paths, output_path. Returns: object
server_infoARead-onlyIdempotent
MapSmith version, licensing, and available engines.
Inputs: no named parameters. Returns: object
simplify_layerAIdempotent
Simplify geometries (Douglas-Peucker, topology preserved) with the drift measured: the manifest records total area and length before and after.
Geographic-CRS inputs are simplified in an estimated UTM zone (decision recorded) and returned in the input CRS — a tolerance in degrees is a different distance at every latitude. On projected CRS the tolerance is interpreted in the CRS units. The feature count is verified unchanged; vertex counts before/after are in the result. Inputs without a CRS are refused.
Inputs: input_path (required), output_path (required), tolerance_meters (required). Returns: object
slopeAIdempotent
Slope gradient from a DEM: GeoTIFF in, GeoTIFF out.
units: degrees (default), percent or radians. DEMs in a geographic CRS are refused — degree cells with meter elevations give plausible but wrong values everywhere; reproject to a projected CRS first. The CRS decision is recorded in the provenance manifest. Requires the [whitebox] extra.
Inputs: dem_path (required), output_path (required), units, z_factor. Returns: object
spatial_joinAIdempotent
Join by spatial predicate (intersects/within/contains).
engine='auto' routes to the fastest available engine for the inputs:
SedonaDB (heavy joins, 10-180x) > DuckDB (GeoParquet fast path) > GeoPandas.
A warnings key in the result flags an empty join or inputs whose extents
do not overlap, each with a hint. Inputs without a CRS are refused.
Inputs: left_path (required), right_path (required), output_path (required), engine, +1 more. Returns: object
validate_planARead-onlyIdempotent
Statically validate a multi-step geoprocessing plan BEFORE running anything.
Write the plan as steps in execution order; each step has a unique id, an operation name from list_operations, and its arguments. Use "$step_id" as an argument value to consume the output dataset of an earlier step. Checks: operations exist and are installed, arguments complete and well-typed, references resolve backwards (mis-ordered steps are rejected), input files exist, outputs don't collide, and CRS flow is simulated end-to-end from the real input files. Returns machine-actionable errors/warnings/notes plus the simulated output CRS per step. Nothing is executed and nothing is written.
Example plan: {"goal": "wells at risk", "steps": [ {"id": "buf", "operation": "buffer_layer", "arguments": {"input_path": "wells.gpkg", "distance_meters": 300, "output_path": "buf.parquet"}}, {"id": "cut", "operation": "clip_layer", "arguments": {"input_path": "$buf", "mask_path": "zone.gpkg", "output_path": "risk.parquet"}}]}
Inputs: plan (required). Returns: object
watershedAIdempotent
Watershed of each pour point: DEM + points in, basin raster out (GeoTIFF).
Basins get 1-based IDs following the pour-point feature order; cells not draining to any point stay nodata. Points are aligned to the DEM CRS automatically (decision recorded). Requires the [whitebox] extra.
Inputs: dem_path (required), output_path (required), pour_points_path (required). Returns: object
zonal_statisticsAIdempotent
Statistics of a raster within each vector zone (exact fractional pixel coverage).
stats: subset of count/sum/mean/median/min/max/stdev/variance/majority/minority/
variety (default: count, mean, min, max). Zones are aligned to the raster CRS
automatically; the decision is recorded in the provenance manifest.
weights_path: a single-band raster of weights on the SAME grid as the values
(same CRS, cells and registration; otherwise refused with how to align it),
for weighted_mean/weighted_sum/weighted_stdev/weighted_variance/weighted_frac --
e.g. mean heat per zone weighted by population (default stats then: count,
mean, weighted_mean). A cell with a value and no weight counts with weight 0,
and the record names the zones where that happened.
Zones without a CRS are refused; warnings and repairs keys in the result
flag a suspicious outcome or geometry MapSmith had to repair.
Requires the [raster] extra.
Inputs: zones_path (required), output_path (required), raster_path (required), stats, +1 more. Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 29 tools
Each operation targets a distinct verb+resource (buffer vs clip vs overlay vs dissolve vs merge), and the descriptions actively pre-empt the classic confusions — merge_layers explicitly says 'this is an append, not a geometric union', and get_provenance explains why lineage is deliberately not a second tool. The one real overlap is structural: run_operation can execute any cataloged operation, so it competes with every exposed tool, though the description frames it as a fallback after list_operations. Terrain tools (slope, aspect, hillshade, flow_accumulation, watershed) share DEM inputs but produce clearly different outputs.
The dominant convention is snake_case verb_noun (buffer_layer, convert_format, validate_plan, describe_dataset, get_provenance), applied consistently across the vector, planning, and provenance families. The deviations are the terrain/hydrology tools, which are bare nouns (slope, aspect, hillshade, watershed, flow_accumulation) with no verb, plus server_info. Still highly readable and predictable, just not one uniform pattern.
29 tools is above the comfortable 3-15 band and the set is grouped into many families (terrain, vector, join, SQL, raster stats, planning, provenance, preview, discovery), which puts real load on selection. The design is self-aware — it caps exposure on purpose and routes the long tail through run_operation — and every tool does map to a distinct operation, so nothing is redundant. Still borderline heavy for what could be a smaller hot-path set.
Coverage of the stated geoprocessing domain is broad and lifecycle-aware: geometry ops, overlay, joins, reprojection, format conversion, terrain/hydrology, zonal raster statistics, SQL, inspection, plus a validate/brief/execute planning triad with provenance. Gaps exist at the tool layer — no raster clip/reproject/resample/calculator, no point-creation or network operations — but run_operation plus list_operations makes the full catalog reachable, so these are workarounds rather than dead ends.
Maintenance
Related MCP Connectors
Paid deterministic data-quality and execution-verification tools for AI agents.
Deterministic reasoning stack for AI agents: simulate, decide & compute, plus cross-domain tools.
Verifiable geospatial tools: distances, topology, isochrones, Overture Maps, with sources.
Geodata for any place, inside your agent: map data, terrain, routes, satellite images, GIS files.
Related MCP Servers
AlicenseBqualityCmaintenanceEnables AI agents to become geospatially intelligent assistants with tools for location search, smart routing, round trip planning, reverse geocoding, isochrone analysis, route visualization, geofence management, and interactive map display.812 npm8Apache 2.0- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to perform geospatial analysis, spatial data processing, vector and raster operations, coordinate transformations, spatial statistics, mapping, and access to climate, satellite, and other geospatial data through MCP-compatible clients.MIT
- AlicenseNot gradedqualityCmaintenanceEnables AI assistants to directly operate ArcGIS Pro through natural language, giving access to 2,500+ spatial analysis tools across 46 toolboxes via the Model Context Protocol.1MIT
- AlicenseNot gradedqualityCmaintenanceEnables AI agents to inspect geospatial vector and raster metadata and run bounded analysis on administrator-registered datasets, including planar buffering, clipping, spatial joins, and PNG map rendering delivered as checksummed artifacts. All data access is read-only and controlled through a curated dataset catalog.Apache 2.0