IfcMCP¶

IfcMCP is an MCP (Model Context Protocol) server that exposes IfcOpenShell query and edit tools to AI coding assistants such as Claude. It wraps IfcQuery and IfcEdit, holding the IFC model in memory across tool calls so no file I/O is required between operations.

The ifcmcp package can also be used directly as a Python library without the MCP server layer.

Installation¶

To use IfcMCP as an MCP server, install it together with the mcp package:

pip install 'ifcmcp[mcp]'

If you only want to use the library directly (without an MCP client):

pip install ifcmcp

Or install from source:

Install IfcOpenShell
Clone the IfcOpenShell repository.
cd /path/to/IfcOpenShell/src/ifcmcp
pip install '.[mcp]'

Setup¶

Add the server to your MCP client. For Claude Code:

claude mcp add --transport stdio ifc -- ifcmcp

Or add to .mcp.json:

{
  "mcpServers": {
    "ifc": {
      "type": "stdio",
      "command": "ifcmcp"
    }
  }
}

Available tools¶

Session tools

ifc_new(schema="IFC4") — create a new empty model in memory
ifc_load(path) — open an IFC file into memory
ifc_reset() — unload the current model, freeing all session state
ifc_save(path="") — write model to disk; empty path overwrites the original

Query tools

ifc_summary() — schema version, entity counts, project metadata
ifc_tree() — full spatial hierarchy
ifc_info(element_id) — deep inspection by step ID
ifc_select(query) — filter elements by IFC class
ifc_relations(element_id, traverse="") — relationships for an element
ifc_clash(element_id, ...) — geometric intersection and clearance checks
ifc_validate(express_rules=False) — schema and constraint validation
ifc_schedule(max_depth=None) — work schedules with nested task trees
ifc_cost(max_depth=None) — cost schedules with nested cost item trees
ifc_schema(entity_type) — IFC class documentation
ifc_contexts() — geometric representation contexts
ifc_materials() — material definitions

Drawing and rendering tools

ifc_plot(...) — generate a 2D drawing via ifcopenshell.draw and return it as an inline image the AI assistant can inspect; SVG always available, PNG requires cairosvg
ifc_render(...) — off-screen 3D render returned as an inline PNG image the AI assistant can inspect; requires pyvista and the IfcOpenShell C++ geometry bindings

ShapeBuilder tools

ifc_shape_list() — list all available ShapeBuilder methods
ifc_shape_docs(method) — documentation for a specific ShapeBuilder method
ifc_shape(method, params="{}") — execute a ShapeBuilder method; entity references resolved by step ID

Edit tools

ifc_list(module="") — list API modules or functions
ifc_docs(function_path) — documentation for an API function
ifc_edit(function_path, params="{}") — execute an ifcopenshell.api mutation
ifc_quantify(rule, selector="") — run quantity take-off; writes IfcElementQuantity psets in-place

Typical workflow¶

ifc_load("/path/to/model.ifc")
ifc_summary()
ifc_tree()
ifc_info(42)
ifc_edit("root.remove_product", '{"product": "42"}')
ifc_save()