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., "@ifc-mcp load the IFC file 'office.ifc' from my desktop"
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.
An MCP (Model Context Protocol) server that enables AI agents to understand and query IFC (Industry Foundation Classes) building model files. This server exposes a rich set of tools for loading IFC files, traversing spatial hierarchies, inspecting elements and properties, querying materials and quantities, and retrieving geometry bounding boxes — allowing AI assistants like Claude to reason about building information models without requiring the user to manually extract data.
pip install ifc-mcp
# or with uv:
uv tool install ifc-mcpClaude Desktop config (no path needed):
{
"mcpServers": {
"ifc": {
"command": "uvx",
"args": ["ifc-mcp"]
}
}
}uv tool upgrade ifc-mcpOr pin to a specific version:
"args": ["ifc-mcp==0.2.0"]Add the tools to your ~/.claude/settings.json so Claude Code never asks for confirmation:
{
"permissions": {
"allow": [
"mcp__ifc__load_ifc_file",
"mcp__ifc__list_loaded_models",
"mcp__ifc__unload_ifc_file",
"mcp__ifc__get_spatial_structure",
"mcp__ifc__get_element_containment",
"mcp__ifc__get_elements_by_type",
"mcp__ifc__get_element_by_id",
"mcp__ifc__search_elements",
"mcp__ifc__get_property_sets",
"mcp__ifc__get_quantities",
"mcp__ifc__get_material",
"mcp__ifc__get_model_statistics",
"mcp__ifc__get_bounding_box",
"mcp__ifc__get_element_placement",
"mcp__ifc__get_element_local_bbox",
"mcp__ifc__get_element_body_mapping",
"mcp__ifc__get_property_sets_detail",
"mcp__ifc__get_element_by_label",
"mcp__ifc__get_representation",
"mcp__ifc__get_elements_batch"
]
}
}Related MCP server: IFCX MCP
Prerequisites: Python 3.11+, uv
git clone <repo>
cd IfcMcp
uv sync
uv run pytest # verify all tests passClaude Desktop config for local dev:
{
"mcpServers": {
"ifc": {
"command": "uv",
"args": ["run", "python", "server.py"],
"cwd": "/absolute/path/to/IfcMcp"
}
}
}On Windows, use double backslashes: "cwd": "C:\\Users\\...\\IfcMcp"
Every tool (except list_loaded_models) requires a model_id — the value returned by load_ifc_file.
Tool | Parameters | Description |
|
| Load an IFC file; returns |
| — | List all models in memory |
|
| Free a loaded model |
|
| Full spatial hierarchy (Project→Site→Building→Storey→Space) |
|
| Full containment chain up to Site |
|
| Paginated list by IFC type |
|
| Full element details (includes |
|
| Look up element by integer STEP label ( |
|
| Batch query; |
|
| Filter elements by type/name/pset/property |
|
| All Psets for an element |
|
| Psets split by instance-level vs type-level |
|
| All quantity sets for an element |
|
| Material info (name, layers, category) |
|
| Element counts by type, storey names, schema |
|
| Min/max XYZ bounding box in world coordinates |
|
|
|
|
| Bounding box in element's local coordinate frame |
|
| Body mapping matrix, world transform, mirroring detection |
|
| Representation structure without tessellating geometry |
Each loaded model exposes a pre-computed summary resource:
ifc://model/{model_id}/summaryThis resource is computed at load time and contains the project name, IFC schema version, element counts by type, and storey names. It gives AI agents a fast overview of a model without issuing individual tool calls.
All tools return structured error objects. Common codes:
{"error": "element_not_found", "details": "..."}
{"error": "model_not_loaded", "details": "..."}
{"error": "geometry_unavailable", "details": "..."}
{"error": "file_not_found", "details": "..."}
{"error": "load_failed", "details": "..."}
{"error": "query_failed", "details": "..."}Get the bounding box (min/max XYZ and dimensions) of an element in project coordinates.
Inputs: model_id (required), global_id (required).
Get the body mapping matrix for an element with IfcMappedItem geometry. Returns: has_mapped_item, body_mapping_matrix (4x4), world_transform (OPM × body mapping), world_transform_determinant, is_mirrored (det<0 means Y-negation needed). Essential for detecting baked rotations in luminaire imports.
Inputs: model_id (required), global_id (required).
Get full details for a single element by its GlobalId.
Inputs: model_id (required), global_id (required).
Look up an element by its integer IFC entity label (#NNN from the STEP file). Returns the same response as get_element_by_id. Use to cross-reference import log output like '[OK] #16535 LightGraphix...' directly against MCP queries.
Inputs: model_id (required), entity_label (required).
Get the spatial containment chain for an element (Space > Storey > Building > Site).
Inputs: model_id (required), global_id (required).
Get the local-space bounding box of an element (no world transform applied). Returns min/max XYZ and dimensions in the element's own coordinate system. Use for mounting axis prediction and recessed fixture detection.
Inputs: model_id (required), global_id (required).
Get the world-space placement of an element: insertion point (XYZ) and orientation axes (X/Y/Z unit vectors). Useful for position, rotation, and aim direction of elements like luminaires.
Inputs: model_id (required), global_id (required).
Batch query for multiple elements by GlobalId. include options: 'entity_label', 'placement', 'property_sets', 'local_bbox'. Default: ['entity_label', 'placement']. Avoids individual round-trips for files with 200+ fixtures.
Inputs: model_id (required), global_ids (required), include.
Get elements of a given IFC type (e.g. IfcWall, IfcDoor). Supports pagination via limit/offset.
Inputs: ifc_type (required), model_id (required), limit, offset.
Get material information for an element. Returns {name, layers, category} — layers is a list of {name, thickness} for layered assemblies, or None for simple materials.
Inputs: model_id (required), global_id (required).
Get element counts by IFC type, storey names, schema version, and total count.
Inputs: model_id (required).
Get all property sets (Psets) for an element as {pset_name: {prop_name: value}}.
Inputs: model_id (required), global_id (required).
Get property sets split by source: instance-level psets vs type-level psets. Use to determine whether a property was overridden at instance level or inherited from the type.
Inputs: model_id (required), global_id (required).
Get all quantity sets for an element as {qset_name: {quantity_name: value}}.
Inputs: model_id (required), global_id (required).
Get the representation structure of an element without tessellating geometry. Returns representation type, identifier, and items with IfcMappedItem details (mapping_target_is_identity, mapping_target_matrix). Use to detect baked rotations before geometry extraction.
Inputs: model_id (required), global_id (required).
Get the full spatial hierarchy: Project > Site > Building > Storey > Space.
Inputs: model_id (required).
List all IFC models currently loaded in memory.
Inputs: no named parameters.
Load an IFC file from the given path. Returns model_id for subsequent queries.
Inputs: path (required).
Search elements with optional filters: IFC type, name substring, pset_name (must own this pset), property_name/property_value (property within that pset or any pset). Supports pagination via limit/offset.
Inputs: model_id (required), limit, offset, ifc_type, +4 more.
Unload an IFC model from memory to free resources.
Inputs: model_id (required).
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 20 tools
Most tools have clearly distinct purposes (loading, spatial structure, element queries, properties, geometry). The only potential confusion is between get_element_by_id and get_element_by_label (both return element details) and get_property_sets vs get_property_sets_detail, but the descriptions explicitly differentiate them by input and level of detail.
All 20 tool names follow a consistent snake_case verb_noun pattern (load_ifc_file, list_loaded_models, get_element_placement, search_elements). The verbs are uniform: load/list/unload for lifecycle, get for queries, search for search, with no mixed conventions or vague action words.
With 20 tools, the server is slightly over the ideal 3-15 range but still well-scoped for the complex domain of IFC model analysis. Each tool addresses a specific aspect of querying building data, and the count is justified by the breadth of information exposed.
The toolset covers the full read-only lifecycle: loading/unloading models, querying spatial hierarchy, elements, properties, quantities, materials, geometry context, and batch operations. Minor gaps exist, such as no direct query of elements by spatial level or retrieval of raw geometry tessellation, but the core analytical workflows are well supported.
Design buildings as IFC4 BIM models from a brief: an AI architect that plans, builds and edits.
Convert Revit files to XKT, IFC, or DWG and query BIM data via natural language.
Check floor-plan images, trace them into editable plans, export PDF and IFC, and query IFC models.
Read, query, edit and render DWG drawings with AI. No AutoCAD required.