Flint
API server fro Blender MCP on macOS with visionOS connectivity. Needs a Claude API key.
claude mcp add --transport stdio create-with-swift-flint python start_server.py \ --env PYTHON_VENV=".venv" \ --env BLENDER_PATH="/Applications/Blender.app/Contents/MacOS/Blender" \ --env EXPORT_FOLDER="exports" \ --env ANTHROPIC_API_KEY="your-anthropic-api-key"
How to use
Flint is a FastAPI-backed MCP server that orchestrates a pipeline to turn text prompts into Blender scenes and export USDZ assets. It exposes endpoints for health checks, running prompt-driven generation, and downloading the latest export. You interact with Flint via the /run and /run-high endpoints to generate content, and the /download endpoint to retrieve the resulting USDZ file. The server leverages the blender-mcp toolchain to drive Blender actions from the language model's outputs, with built-in context guards and tool allowlists to keep prompts reliable and within resource limits. To begin, ensure the environment is prepared (Blender installed at the expected path, an Anthropic API key, and an active virtual environment), then start the Python server and call the endpoints with a JSON payload containing your prompt.
Once running, you can use:
- POST /run for standard-quality MCP generation (token-safe, fewer steps).
- POST /run-high for high-detail MCP generation (more steps and richer hints).
- GET /health to verify Blender, Claude integration, and export folder readiness.
- GET /download/{filename} to fetch the latest USDZ export. The server will save assets to the configured EXPORT_FOLDER (defaults to the project root's exports directory) and ensure both the .blend and .usdz artifacts are produced for downstream consumption.
How to install
Prerequisites
- macOS with Blender installed and accessible at /Applications/Blender.app/Contents/MacOS/Blender (or update BLENDER_PATH accordingly)
- Python 3.10+
- uv toolchain for MCP integration (uvx blender-mcp) -> install with: brew install uv
- Optional: a Python virtual environment for isolation
Step-by-step setup
- Install Blender and Python prerequisites
- Ensure Blender is reachable at the path documented above. If not, adjust the BLENDER_PATH in .env or in the code.
- Create and activate a virtual environment
- python -m venv .venv
- source .venv/bin/activate
- Install Python dependencies
- pip install -r requirements.txt
- Prepare environment configuration
- Create a .env file in the project root and set at minimum: ANTHROPIC_API_KEY=sk-your-key BLENDER_PATH=/Applications/Blender.app/Contents/MacOS/Blender # override if needed EXPORT_FOLDER=exports # optional; defaults to project root
- Optional defaults file
- You can copy config.env to .env to bring in shared defaults, then override as needed: cp config.env .env
- Start the server
- python start_server.py # hot reload is enabled by default in the project script
- Verify the service
- curl http://localhost:8000/health
- You should see a healthy status when Blender, the Claude API, and the export folder are accessible.
Additional notes
Tips and notes:
- Environmental variables defined in .env (or via the mcp_config env block) must be available to both FastAPI and the export script for proper operation.
- Ensure EXPORT_FOLDER is writable; otherwise, exports may fail or be saved to an unexpected location.
- If you encounter prompt-too-long errors, switch to /run for standard profiles or shorten the prompt or reduce the MCP allowlist.
- The integration leverages Claude Sonnet 4.5 with the blender-mcp toolchain to guide Blender operations; ensure your Anthropic API key remains valid to avoid rate limits.
- If Blender path errors occur, confirm BLENDER_PATH matches your installed Blender binary or update the path accordingly.
- You can customize the export path or file naming by adjusting EXPORT_FOLDER and the underlying export_model.py logic.
Related MCP Servers
osaurus
AI edge infrastructure for macOS. Run local or cloud models, share tools across apps via MCP, and power AI workflows with a native, always-on runtime.
swift -gui
MCP server that can execute commands such as keyboard input and mouse movement on macOS
mcp-client-example
Learn how to implement MCP client with SwiftUI and Anthropic API
mirroir
MCP server for controlling a real iPhone via macOS iPhone Mirroring...and any MacOs app. Screenshot, tap, swipe, type — from any MCP client.
claude-command-runner
Swift-based MCP server that bridges Claude Desktop with terminal applications, enabling seamless command execution with intelligent output retrieval. Features auto-capture, SQLite history, and Warp Terminal integration
swift
Full-featured Swift SDK for Model Context Protocol servers and clients