prosuite-mcp
OfficialThis server acts as an MCP gateway for Dira ProSuite geospatial quality verification, enabling AI assistants to perform automated GIS data quality checks through natural language interaction.
Load and manage QA specs: Load a
.qa.xmlspec file at runtime (load_spec), inspect its contents including specifications, workspace definitions, and datasets (describe_spec).Search and discover conditions: Search the loaded spec using natural language queries in English, German, French, or Italian (
search_spec), or list all available quality conditions filtered by keyword (list_conditions).Inspect conditions: Retrieve detailed documentation for specific quality conditions, including parameter types and whether they expect dataset names or primitive values (
describe_condition).Run quality verifications: Execute a named
QualitySpecificationfrom a loaded XML spec with workspace path substitutions (run_xml_verification), or perform ad-hoc verifications by specifying a workspace, datasets, and conditions directly (run_verification).Preview and test: Preview the XML representation of a condition (
condition_to_xml), see how adding a condition would modify a spec (add_condition_to_spec), and preview the execution results of a proposed condition before committing (preview_condition_run).Spatial and output filtering: Optionally restrict verifications to a bounding box and direct output (Issues.gdb, HTML report) to a server-side directory.
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., "@prosuite-mcp Check road connectivity in C:/data/tlm.sde"
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.
prosuite-mcp
MCP server that exposes Dira ProSuite quality verification to AI assistants (Claude, etc.).
Prerequisites
A running ProSuite Quality Verification Server reachable from the host where this server runs.
Related MCP server: QGIS MCP
Configuration
Environment variable | Default | Description |
|
| ProSuite service host |
|
| ProSuite service port |
| (none) | Path to PEM certificate for TLS |
Usage
Windows users: start at docs/windows-setup.md.
CLI coding agents (Claude Code, Copilot CLI, opencode): see docs/cli-clients.md
Any other MCP client (Claude Desktop, other GUI apps, or anything else): see docs/gui-clients.md
No coding agent permitted? An experimental example drives the same tools from any OpenAI-compatible LLM, local or hosted: see docs/local-llm.md
Tools
Every tool is defined in src/prosuite_mcp/tools.py: its docstring is the authoritative description, the same one your MCP client shows the LLM.
Example
Once connected, you talk to the assistant in plain language:
Check road connectivity in
C:/data/tlm.sde.
With a spec loaded, the assistant calls describe_spec to see which specifications and workspaces it defines, then run_xml_verification to run one against your data, and returns a summary of errors per condition. The spec goes to the ProSuite service as-is, so per-condition filters and defaults are applied exactly as your domain experts configured them. search_spec browses the conditions in a spec by keyword.
list_datasets and describe_dataset read the geodatabase itself, so the assistant can pick conditions from the actual geometry type, feature count and field names rather than from a dataset's name. These read the path on the machine running this server, which need not be the one the ProSuite service resolves paths on.
list_conditions and describe_condition are enriched with the ProSuite QA Quick Reference: each test gains the readable description and the test family it belongs to, and searches match that wording rather than only the API docstring. The document is downloaded in the background when the server starts, and a lookup never waits for it, so calls made before it lands, or on a host that cannot reach it, return the same output as before without the enrichment.
Without a spec, the assistant uses list_conditions and describe_condition to find and configure conditions from scratch, then run_verification to run them ad-hoc. Datasets take a filter_expression here too, but only one per dataset per run, so a spec that filters the same feature class differently in two conditions cannot be reproduced this way.
Development
See docs/development.md.
Available Tools
9 toolsadd_condition_to_specA
Preview adding a new QualityCondition to a spec, reusing an existing descriptor.
Builds the condition through the same prosuite factory as run_verification, resolves a matching (never synthesizes one), and returns the full updated spec XML with the condition appended and wired into target_specification_name. Preview only: never writes to a file. Call describe_spec first for valid specification/workspace/dataset names.
Args: target_specification_name: QualitySpecification to wire the condition into. name: Human-readable condition name; must not already exist in the spec. condition_request: {condition: method name from list_conditions, params: dict}. datasets: Feature classes/tables used by condition_request, each with 'name' and an optional 'filter_expression'. workspace_id: Logical workspace id to bind datasets to (e.g. "DATA_OSM"). allow_errors: Whether issues from this condition are tolerated. description: Optional description element. spec_xml: Spec XML text; defaults to reading the active spec (the one load_spec was last called with, otherwise PROSUITE_SPEC_PATH). category: Which category to file the condition under. Only needed when the spec nests QualityConditions under more than one.
Returns 'status': 'ok' with 'spec_xml' holding the updated spec, ready to review and persist yourself, or 'status': 'error'.
Inputs: name (required), datasets (required), workspace_id (required), condition_request (required), +5 more. Returns: object
describe_conditionA
Describe the parameters of a ProSuite quality condition.
Returns the full docstring and parameter list with types. Dataset parameters expect a dataset name string (must match a name in the datasets list you will pass to run_verification). Primitive parameters take their direct value (number, bool, string).
Parameters marked optional may be left out of run_verification's params: ProSuite applies its own default. Everything else must be supplied.
Inputs: name (required). Returns: result
describe_specA
Describe the loaded QA spec file: available specifications, workspace definitions, and per-specification summary of conditions, workspace IDs, and dataset names.
Call this before run_xml_verification to learn:
Which specification_name values exist in the spec (pass one to run_xml_verification)
Which workspace_id values need to be replaced with real paths
Which datasets each specification expects (useful for sanity-checking the workspace)
Describes whichever spec is active: the one load_spec was last called with, otherwise PROSUITE_SPEC_PATH. Returns 'status': 'ok' or 'error'.
Inputs: no named parameters.
list_conditionsA
List available ProSuite quality conditions.
Returns condition names and their one-line description. Use search to filter by keyword (matched against name and description). Pass a result name to describe_condition to get full parameter details before building a run_verification call.
Inputs: search. Returns: result
load_specA
Load a .qa.xml spec file at runtime.
Replaces any previously loaded spec, and takes precedence over PROSUITE_SPEC_PATH, so every subsequent spec tool (search_spec, describe_spec, add_condition_to_spec, run_xml_verification) uses the new file. Use this when the spec path is only known at conversation time (e.g. a file on OneDrive or a network share) instead of pre-configuring PROSUITE_SPEC_PATH.
Args: path: Absolute path to the .qa.xml spec file on the local machine.
Returns 'status': 'ok' with 'conditions_loaded', or 'status': 'error'.
Inputs: path (required).
preview_condition_runA
Run a single proposed condition ad-hoc and show what it actually flags.
Tier-2 confirmation for spec authoring: add_condition_to_spec only confirms a condition builds and references an existing descriptor. This runs it for real against model_catalog_path and returns the same engine-confirmed summary as run_verification (engine_confirmed, total_errors, sample_features with the actual flagged issues), so a proposed condition can be judged by what it flags before it's merged into a spec. Scope envelope to a small extent to keep this a preview, not a full run.
Args: model_catalog_path: Workspace path on the server, e.g. 'C:/data/mydb.gdb' or a .sde connection file. condition_request: {condition: method name from list_conditions, params: dict}. datasets: Feature classes/tables used by condition_request, each with 'name' and an optional 'filter_expression'. workspace_id: Logical name for the data model (arbitrary, used in generated condition names). output_dir: Optional server-side directory for Issues.gdb and HTML report. Omitted, a local runs/ directory is used only when PROSUITE_HOST is local, since the service resolves the path. envelope: Optional spatial filter {x_min, y_min, x_max, y_max}.
Inputs: datasets (required), workspace_id (required), condition_request (required), model_catalog_path (required), +2 more. Returns: object
run_verificationA
Run a ProSuite quality verification.
Build an ad-hoc condition-list specification and run it against the given workspace. The ProSuite service (prosuite-qa-microservice) must be reachable at the host/port configured via PROSUITE_HOST / PROSUITE_PORT environment variables (default: localhost:5151).
Args: model_catalog_path: Workspace path on the server, e.g. 'C:/data/mydb.gdb' or a .sde connection file. model_name: Logical name for the data model (arbitrary, used in generated condition names). datasets: Feature classes or tables to make available for conditions. Each entry has a 'name' (feature class name) and an optional 'filter_expression' (SQL WHERE clause). conditions: Conditions to run. Each entry has: - condition: method name from list_conditions (e.g. 'qa_min_length_0') - params: dict mapping parameter names to values. Dataset parameters take a string matching a name in 'datasets'; primitive parameters take their direct value. output_dir: Optional server-side directory for Issues.gdb and HTML report; the service process must have write access. Omitted, a local runs/ directory is used only when PROSUITE_HOST is local, since the path is resolved on the service's machine. envelope: Optional spatial filter {x_min, y_min, x_max, y_max}. Omit for full-extent verification.
Returns a summary with status, total_errors, and per-condition breakdown. Check 'status': 'error' for connection or parameter failures.
Inputs: datasets (required), conditions (required), model_name (required), model_catalog_path (required), +2 more. Returns: object
run_xml_verificationA
Run a ProSuite quality verification directly from the loaded XML spec file.
Unlike run_verification, this tool sends the XML spec to the ProSuite service as-is, without decomposing it into individual conditions and datasets. This preserves per-condition dataset filters, default scalar values, and all other spec details exactly as configured.
Use search_spec (with empty query) to discover available specification_name values and workspace_id keys that need to be replaced.
Args: specification_name: Name of the QualitySpecification element inside the XML file to run (e.g. 'Copy of DATA_OSM_10_Demo'). data_source_replacements: Maps each workspace_id in the XML to the actual workspace path on the ProSuite server. Example: [{"workspace_id": "DATA_OSM", "workspace_path": "C:/data/osm.sde"}] output_dir: Optional server-side directory for Issues.gdb and HTML report. Omitted, a local runs/ directory is used only when PROSUITE_HOST is local, since the service resolves the path. envelope: Optional spatial filter {x_min, y_min, x_max, y_max}.
Returns a summary with status, total_errors, and per-condition breakdown. Runs whichever spec is active: the one load_spec was last called with, otherwise PROSUITE_SPEC_PATH.
Inputs: specification_name (required), data_source_replacements (required), envelope, output_dir. Returns: object
search_specA
Search the loaded QA spec for conditions matching a natural-language query.
Returns up to max_results conditions whose name, description, or category contains the query string (case-insensitive, literal substring match, not semantic). Spec content may be in English, German, French, or Italian -- query in the same language as the spec to find matches.
Each result includes:
name: the full condition name (human-readable rule statement)
category: domain grouping from the spec
allow_errors: False means a hard failure, True means tolerated
condition_request: ready to pass directly into run_verification's conditions list (includes condition method name and pre-filled params)
required_datasets: dataset names and filter expressions to include in run_verification's datasets list
A result carrying 'unsupported': True has neither of those last two: the programmatic path cannot rebuild it (transformer preprocessing, or no factory method). It is listed so browsing sees the whole spec, and it still runs normally under run_xml_verification, which reads the spec as written.
Searches whichever spec is active: the one load_spec was last called with, otherwise PROSUITE_SPEC_PATH. Returns 'status': 'ok' or 'error'.
Inputs: query (required), max_results.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 9 tools
Most tools have clearly distinct purposes: listing/describing conditions versus describing/searching/loading specs versus running verifications. However, run_verification and run_xml_verification both execute verifications and could be confused if an agent doesn't carefully read the parameter differences, and preview_condition_run partially overlaps with run_verification.
All tool names follow a consistent snake_case verb_noun pattern: list_conditions, describe_condition, describe_spec, search_spec, load_spec, add_condition_to_spec, run_verification, preview_condition_run, run_xml_verification. The only slight outlier is preview_condition_run, but it still fits the pattern well.
With 9 tools, the server is well-scoped for its purpose. Each tool addresses a distinct stage in the QA workflow—discovery, spec management, ad-hoc running, and XML-based running—without unnecessary bulk or gaps that would require extra tools.
The toolset covers the core lifecycle: listing conditions, describing them, loading/describing/searching specs, previewing condition additions, and running verifications (both ad-hoc and XML). The main gaps are the lack of a persistent save/update operation for specs (add_condition_to_spec only previews) and no tool for retrieving past verification results or reports.
Maintenance
Related MCP Connectors
Geospatial AI MCP server — satellite imagery, embeddings, weather, GNS governance
MCP server that lets AI assistants use all OneSchema features exposed via the public API.
Hosted MCP server for pre-action AI governance verdicts, model routing, and evidence verification.
41Official Microsoft MCP Server to query Microsoft Entra data using natural language
Related MCP Servers
- AlicenseAqualityDmaintenanceA Model Context Protocol server that connects Large Language Models to the GeoServer REST API, enabling AI assistants to query and manipulate geospatial data through natural language.997MIT
- AlicenseBqualityAmaintenanceAn MCP server that enables AI assistants to directly control QGIS for tasks like layer management, feature editing, and map rendering. It provides a suite of 50 tools to execute processing algorithms and manage GIS projects through natural language commands.125447GPL 2.0
- AlicenseNot gradedqualityDmaintenanceMCP server for interacting with Google Earth Engine, enabling geospatial analysis such as dataset visualization, statistics computation, and search via AI assistants.16MIT
- AlicenseNot gradedqualityDmaintenanceA Model Context Protocol (MCP) server that exposes GeoServer REST API functionality for natural language interaction through AI assistants like Claude, VS Code Copilot, and other MCP-compatible clients.47 PyPIBSD 2-Clause "Simplified"