Autodesk Inventor MCP
Connect AI assistants to a live Autodesk Inventor session over COM: inspect bodies and features, find sliver faces in imported models, dry-run Unwrap, and script the Inventor API.
Que pouvez-vous faire avec Autodesk Inventor MCP ?
- agent-receipts: Skills that make AI coding agents prove "done" with real test output
- blender-dlss5-neural-rendering: Blender viewport and renders through DLSS 5 neural rendering
- metahuman-face-capture: MetaHuman face capture from a webcam in Blender
- unreal-groom-alembic-exporter: Export UE Groom assets (MetaHuman hair) to Alembic
Installation
Les étapes d'installation sont dans la documentation ci-dessous. Documentation
Documentation
Tirée du README du projet sur GitHub, actualisée chaque semaine. Les liens ouvrent GitHub.
Sommaire
inventor-mcp
An MCP server that gives AI agents (Claude Code, Claude Desktop, Cursor, Codex…) a live connection to Autodesk Inventor.
Ask your agent things like "which faces in this imported STEP are tiny slivers?" or "will Unwrap work on these 12 faces?" and it reads the answer straight from the open model.
- Zero dependencies: one Python file, standard library only.
- Live: talks to the running Inventor session over COM (via Windows PowerShell). No export, no file round-trips.
- Units you think in: lengths in mm, areas in mm² (Inventor's internal unit is cm; the server converts).
- Safe by default:
unwrap_trycreates the feature, measures the result and deletes it again unless you passkeep: true.
Requirements
- Windows with Autodesk Inventor installed and running (developed against Inventor 2027)
- Python 3.9+
Install
Clone or download server.py, then register it with your agent.
Claude Code
claude mcp add inventor -- python C:\path\to\inventor-mcp\server.py
Claude Desktop / Cursor / any MCP client (mcpServers config):
{
"mcpServers": {
"inventor": {
"command": "python",
"args": ["C:\\path\\to\\inventor-mcp\\server.py"]
}
}
}
Tools
| Tool | What it does |
|---|---|
status | Inventor version and open documents |
model_info | Bodies (face/edge/vertex counts, bounding box in mm) and the feature tree of a part |
smallest_faces | The N smallest faces with area and center. Finds import defects fast |
face_neighbors | Faces that share an edge with a given face |
unwrap_try | Tries an Unwrap on a face set and reports success, timing, result area and bounding box, the details the GUI doesn't show. Rolls back by default |
script | Runs arbitrary PowerShell against Inventor's COM API, with $inv and helpers predefined |
Face indices are 1-based and stable for a given body as long as the model doesn't change.
⚠️ About the script tool
script executes any PowerShell your agent writes, with your user's permissions. That's what makes it powerful (anything in the Inventor API is reachable), and it's also why you should keep your agent's tool-approval prompts on for it. If you don't want it, delete its entry from TOOLS in server.py.
All other tools pass arguments as data, never as code.
Example (illustrative)
You: The STEP I imported won't unwrap. What's wrong with it?
Agent: (calls
smallest_faces) Body 1 has 412 faces; the 6 smallest are under 0.01 mm², clustered around (12.4, -3.1, 40.0). (callsface_neighborson face 287) They sit between two large B-spline faces, which is the classic sliver from a bad tessellation export. (callsunwrap_trywithout them) Unwrap succeeds in 1.8 s when those 6 faces are excluded.
Contributing
Issues and PRs are welcome, especially new read-only tools (assemblies, sketches, iProperties, sheet metal), Inventor-version compatibility reports, and non-English Windows fixes. See the issues labeled good first issue.
License
MIT. Not affiliated with or endorsed by Autodesk. Autodesk and Inventor are trademarks of Autodesk, Inc.
More tools by effectustasi
- agent-receipts: Skills that make AI coding agents prove "done" with real test output
- blender-dlss5-neural-rendering: Blender viewport and renders through DLSS 5 neural rendering
- metahuman-face-capture: MetaHuman face capture from a webcam in Blender
- unreal-groom-alembic-exporter: Export UE Groom assets (MetaHuman hair) to Alembic
Questions fréquentes
Qu'est-ce que Autodesk Inventor MCP ?
Connect AI assistants to a live Autodesk Inventor session over COM: inspect bodies and features, find sliver faces in imported models, dry-run Unwrap, and script the Inventor API.
Comment installer Autodesk Inventor MCP ?
Suivez les étapes d'installation de sa documentation, affichée sur cette page et dans son dépôt.
Autodesk Inventor MCP est-il open source ?
Oui. Son code est sur GitHub sous licence MIT.
Qui maintient Autodesk Inventor MCP ?
Il est publié par effectustasi. AI.tn le référence sans le maintenir ni le recommander.
Issu des listes communautaires awesome-mcp-servers et du dépôt GitHub du projet. Être listé n'est pas une recommandation : vérifiez le code et les permissions d'un serveur avant de lui donner accès à vos données.