Give AI full read-write access to Autodesk Revit.
RevitMCPBridge
<strong>Give AI full read-write access to Autodesk Revit. 705+ endpoints. Zero manual steps.</strong>
Quick Start ·
What It Enables ·
API Reference ·
Autonomy Levels
Coming from revit-mcp? That project was archived on Feb 26, 2026. RevitMCPBridge2026 is the actively maintained alternative with 705+ endpoints, named pipe communication (no crashes), and a 113-file architectural knowledge base. See Quick Start to get running in minutes.
What is this?
RevitMCPBridge is a Revit add-in that exposes the entire Revit API through the Model Context Protocol (MCP) via named pipes. AI systems — Claude, GPT, custom agents — can read, create, modify, and validate anything in a Revit model programmatically. No Dynamo. No manual steps. Just send JSON, get results.
For AEC professionals: Your AI assistant can now open your Revit model and actually do things — create walls, place doors, generate sheets, check code compliance, produce construction documents.
For developers: 705+ typed endpoints with parameter validation, transaction management, and structured error responses. Connect any language that can write to a named pipe.
Why Named Pipes Instead of HTTP?
Revit is single-threaded. Every API call must execute on the main UI thread.
| Approach | What Happens Under Load |
|---|---|
| HTTP server | Competes for the main thread. Multiple concurrent AI requests cause timeouts, dropped connections, and crashes. |
| Named pipes | Uses Revit's own ExternalEvent queue. Requests are serialized through the same mechanism Revit uses internally. Zero thread contention, zero crashes. |
Every other Revit MCP implementation uses HTTP. They work for simple demos but break under real workloads. Named pipes are why this bridge can handle 705+ endpoints reliably.
Before and After
Before RevitMCPBridge:
You: "Create a sheet set for these 12 floor plans"
AI: "Here are the steps you should follow manually in Revit:
1. Go to View > Sheet Composition > New Sheet
2. Select your title block...
3. Repeat 12 times..."
After RevitMCPBridge:
# AI executes directly in Revit
for i, view_id in enumerate(floor_plan_ids):
call_revit("createSheet", {
"sheetNumber": f"A1.{i+1}",
"sheetName": f"Floor Plan - Level {i+1}",
"titleBlockName": "E1 30x42 Horizontal"
})
call_revit("placeViewOnSheet", {
"sheetId": sheet_id,
"viewId": view_id,
"locationX": 1.5,
"locationY": 1.0
})
# 12 sheets created, views placed, titled — in seconds
What It Enables
| Capability | Methods | Example |
|---|---|---|
| Read any model data | Parameters, elements, geometry, schedules | Extract every door schedule to JSON |
| Create elements | Walls, doors, windows, rooms, structural, MEP | Build a floor plan from a PDF specification |
| Modify elements | Move, resize, reparameter, retype | Batch-update 200 door fire ratings |
| Generate documents | Sheets, views, schedules, annotations | Produce a full CD set automatically |
| Validate models | Code compliance, clash detection, QC | Check egress paths against IBC requirements |
| AI autonomy | Goal execution, learning, self-healing | "Set up this project" → 360 sheets, done |
Scale
- 705+ MCP endpoints across 25+ categories
- 146 C# source files, 13,000+ lines
- 113 knowledge files of architectural domain expertise (building codes, room standards, MEP systems, material specs)
- 68 unit tests with NUnit
- 5 levels of autonomy — from direct API calls to autonomous g
Tools (4)
create_elementCreate architectural elements like walls, doors, or windows in the Revit model.modify_elementUpdate properties of existing elements such as moving, resizing, or changing types.generate_sheetCreate a new sheet and place specified views onto it.read_model_dataExtract parameters, geometry, or schedule data from the Revit model.Configuration
{"mcpServers": {"revit": {"command": "path/to/RevitMCPBridge.exe"}}}