Revit MCP

What Revit MCP does

Revit MCP lets an MCP-compatible AI client send natural-language requests to Autodesk Revit. The available workflows cover both model changes and project analysis, including wall dimensioning, sheet and viewport layout, view-template assignment, material updates, room settings, scaffold takeoff, clash-related review, and regulatory checks.

The tool set includes domain-specific operations such as analyzing beam penetrations, corridor widths, floor slopes, smoke-detector placement, tall partition rooms, and sanitary fixture requirements. It also includes drafting and documentation operations: arranging drafting-view viewports, aligning titleblocks, adjusting section datums, creating wall sections, applying panel patterns, and renumbering sheets. Several tools support preview or dry-run behavior before changes are committed.

How it works

The architecture has three parts. An AI client launches the Node.js/TypeScript MCP process over stdio. That process forwards requests over a local WebSocket connection, normally at ws://localhost:8964, to the C# Revit add-in. The add-in executes the corresponding Revit API workflow inside Autodesk Revit.

Revit must be open with the MCP service enabled from its ribbon before requests can reach the model. The Revit-side service permits one connected AI client at a time. To change clients, release the current connection from the Revit ribbon, then start or reconnect the next client.

Setup and configuration

The supported environment is Windows 10 or later with Autodesk Revit 2022, 2023, 2024, 2025, or 2026. Revit 2022–2024 use .NET Framework 4.8; Revit 2025–2026 use .NET 8. Node.js LTS, preferably version 20 or later, is also required.

The npm bridge can be started without cloning the repository:

{
  "mcpServers": {
    "revit-mcp": {
      "command": "npx",
      "args": ["-y", "@shuotao/revit-mcp-server"]
    }
  }
}

The C# add-in still needs to be built and deployed for the installed Revit version. Alternatively, the project’s setup script checks prerequisites, installs dependencies, builds the MCP server and add-in, deploys the add-in, and helps configure common clients. Manual builds use version-specific Release configurations, followed by deployment to the matching Revit Addins directory.

For a source build, AI clients can launch MCP-Server/build/index.js with Node. Configuration examples are supplied for Claude Code, Claude Desktop, Gemini CLI, and VS Code Copilot.

Tools and capabilities

• Modify Revit element types, room height parameters, materials, filled regions, curtain-wall panel patterns, and pipe-end fittings.
• Create dimensions, wall sections, crop-box alignments, section datum adjustments, and viewport arrangements.
• Batch-update view templates, titleblock positions, sheet numbers, and viewport documentation.
• Analyze geometry and model conditions for penetrations, slopes, corridor widths, openings, stairs, rooms, detectors, and partitions.
• Perform selected building-code checks, including exterior openings, effective floor openings, smoke-exhaust windows, stair headroom, and supported sanitary-fixture rules.
• Calculate exterior, room-based, and manually traced scaffold perimeters.

Limitations and notes

Only one AI client can hold the Revit-side connection at a time; clients must be switched rather than used concurrently. The Revit add-in is a separate installation from the npm stdio bridge. Some workflows have specific scope: drafting-view viewport arrangement is limited to DraftingView, sanitary-fixture checking currently covers C-1 factories and warehouses, and several analysis tools depend on the active view, selected elements, linked models, or specified Revit element IDs.