CAD2GIS
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| DEEPSEEK_API_KEY | No | API key for the DeepSeek provider used by constrained onboarding/auto-convert. | |
| CAD2GIS_PROJECT_ROOTS | No | Semicolon-separated list of additional authorized root directories for MCP file access outside the client workspace. Usually no configuration is needed. | |
| CAD2GIS_READER_BACKEND | No | Set to "autocad" to use the AutoCAD reader on Windows; otherwise the default reader (LibreDWG) is used. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| get_cad_bundle_statusA | Inspect the optional pinned CAD sidecar without downloading or starting it. |
| prepare_cad_bundle_workspaceC | Create a hash-bound DXF editing copy inside the authorized project roots. |
| list_cad_bundle_toolsC | Discover only tested CAD tools and their path-free argument schemas. |
| call_cad_bundle_toolC | Call an allowed CAD tool on a workspace copy, preserving source and old revisions. |
| audit_visual_conversionC | Audit source-bound geometry, fields and labels with the canonical visual audit. |
| build_qgis_reviewB | Build a new offline QGIS review; preserve matched or uncorrected identity. Uses the canonical QGIS subprocess. An absent registration stays explicit; it is never interpreted as successful matching. Existing SVG bindings are preserved, while new semantic symbol assignments are not invented. |
| build_source_svg_deliveryB | Apply hash-bound source-instance SVGs to a new canonical review package. This preserves CAD database bytes and produces candidate review projects; it does not accept a semantic interpretation or an OSM matching revision. |
| publish_matched_deliveryB | Publish matched scopes as six-file SVG/OSM deliveries using frozen snapshots. |
| verify_qgis_reviewC | Independently reopen and render a hash-bound review QGZ in QGIS. |
| start_review_serverC | Start and verify an owned loopback review service; this does not open a browser. |
| get_review_server_statusC | Verify a service owned by this MCP process; never discover or adopt another listener. |
| stop_review_serverB | Stop only the subprocess owned by this MCP session; never accept an arbitrary PID. |
| get_capabilitiesC | Describe the stable CAD2GIS protocol and accuracy boundaries. |
| debug_mcpC | Inspect identity/runtime, or artifacts/query for an EXISTING run; project config directories are not run_dir. |
| export_sourceC | Create a new immutable CAD snapshot; omit CRS unless authoritative evidence exists. |
| query_source_entitiesC | Query a snapshot using bounded SQL filters, stable keyset cursors and exact facts. |
| get_entity_context_batchC | Read bounded source or instance facts for observed IDs without changing them. |
| prepare_semantic_batchesC | Materialize source-bound semantic candidates in a separate new directory. |
| query_relationship_candidatesC | Read only observed class, label or dimension candidates with registered policy IDs. |
| initialize_semantic_storeC | Initialize a durable revision ledger bound to a verified source and candidate set. |
| inspect_semantic_storeC | Read the authoritative semantic revision and optional durable compile job state. |
| preview_semantic_patchB | Validate an ID-only semantic patch against frozen facts and its base revision. |
| commit_semantic_patchB | Atomically commit a validated semantic patch with revision CAS and idempotency. |
| compile_semantic_revisionC | Compile an immutable source-coordinate semantic candidate, without auto promotion. |
| cancel_compile_jobB | Durably cancel a compile job and fence its worker from publishing a result. |
| reconcile_compile_jobsB | Recover interrupted compile jobs; call only when their workers are known stopped. |
| get_runtime_statusC | Report portable reader and GIS runtime readiness. |
| install_runtimeB | Install the checksum-pinned portable LibreDWG runtime. |
| inspect_sourceB | Read and summarize a new DWG without reusing another source's mapping. |
| bootstrap_projectB | Create a source-bound draft project pack for one previously inspected DWG. |
| validate_projectC | Validate source, inventory, mapping, unit, CRS, and review admission gates. |
| prepare_ai_onboardingC | Return source-bound evidence and a strict proposal schema to the host AI. |
| apply_ai_onboardingC | Compile a source-bound proposal; conversion must use the same geometry_repairs mode. |
| auto_onboard_and_convertC | Run provider-backed onboarding, deterministic admission, and conversion. |
| inspect_runB | Return the immutable manifest summary for one conversion run. |
| audit_runC | Verify one immutable run's artifacts and delivered layer census. |
| package_qgis_deliveryA | Create verified QGZ projects and a complete portable ZIP from an existing run. |
| list_evidence_nodesC | Page through source-bound evidence nodes without changing any facts. |
| get_evidence_nodeC | Read one exact evidence node, including immutable reader facts. |
| list_visual_regionsC | List multi-scale render and hit-map artifacts bound to the graph. |
| resolve_visual_hitC | Resolve one six-digit RGB hit-map value to its source entity node. |
| list_registered_operationsB | Describe the only repair operations a model may propose. |
| list_endpoint_join_candidatesC | List source-derived endpoint pair IDs selectable by a decision pack. |
| list_network_repair_candidatesC | List graph-bound crossing and collinear candidates for derived repairs. |
| validate_decision_packC | Verify source binding, content addresses, and all graph references. |
| create_decision_packC | Create a typed pack from ID-only proposals and validate it locally. |
| decide_iteration_candidateC | Accept/reject/revise a gated candidate; accept needs user confirmation. |
| evaluate_iteration_candidateC | Gate a new run against the active run without promoting it. |
| export_iteration_learningC | Export accepted lessons as source-bound onboarding suggestions. |
| get_cad_scene_nodeC | Read one exact pre-semantic node with its source-bound facts. |
| inspect_iterationB | Inspect iteration state, budget, pending decisions, and the next action. |
| list_scene_visual_regionsC | Page through layout-aware visual regions bound to a CAD Scene Graph. |
| get_scene_visual_region_contextC | Read one visual region's CAD facts without sending all drawing nodes. |
| create_scene_interpretation_planC | Bind host-AI scene-role rankings to immutable graph and visual hashes. |
| validate_scene_interpretation_planC | Validate a persisted AI scene plan against current immutable evidence. |
| list_cad_scene_nodesC | Page through immutable pre-semantic CAD scene nodes. |
| list_label_candidatesC | List advisory, distance-ranked labels without changing source facts. |
| list_legend_catalog_candidatesC | List advisory legend/symbol-sample groups from source-bound facts. |
| prepare_iteration_contextC | Create the constrained evidence/routing pack for the next retry. |
| record_iteration_feedbackC | Bind user language and visual evidence to the active source/run hashes. |
| run_conversionD | Run the canonical pipeline; no conversion logic is duplicated here. |
| start_feedback_iterationB | Start a bounded, auditable retry loop over one immutable conversion run. |
| prepare_review_workspaceC | Create the separate edit store and return a local review launch command. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 63 tools
Many tools target clearly distinct pipeline stages, but several clusters overlap heavily: inspect_semantic_store/debug_mcp/inspect_run/inspect_iteration, build_qgis_review/verify_qgis_review/package_qgis_delivery/prepare_review_workspace, and evaluate_iteration_candidate/decide_iteration_candidate. The detailed descriptions help, but with 63 tools the risk of misselection remains substantial.
Names mostly follow a predictable snake_case verb_noun pattern (inspect_source, validate_project, create_decision_pack, run_conversion). There are minor deviations such as auto_onboard_and_convert and some noun-heavy names, but the overall convention is readable and consistent.
Sixty-three tools is an extreme mismatch for a single MCP server surface, even for a complex CAD-to-GIS pipeline. The set is far too large for an agent to navigate efficiently, and many tools appear to earn their place only inside a very specialized stage.
The surface covers a broad lifecycle: runtime setup, source export/inspection, semantic revision, conversion, audit, QGIS packaging, review serving, and iteration feedback. Some cleanup/deletion or simpler common operations may be missing, but core workflows appear well covered.