Installation¶
Install and run the UML-MCP server locally.
System requirements¶
- Python 3.12 (see
requires-pythoninpyproject.toml) - uv (recommended), Poetry, or pip
- Optional: Docker for local PlantUML or Kroki
Installation steps¶
- Clone the repository:
- Install dependencies:
With uv (recommended):
With Poetry:
With pip:
Alternatively, for a lockfile-derived pip install:
- For development (tests, linting, docs):
Run the server¶
From the project root:
Default transport is stdio (for MCP clients). For local HTTP:
With uv: uv run python server.py (same flags).
Verifying installation¶
You should see a table listing generate_uml and validate_uml, plus registered prompts/resources in logs or --list-tools output depending on UI settings.
IDE integration¶
Point your MCP client at server.py with absolute paths for args and cwd (project root); optional MCP_OUTPUT_DIR in env. config/README.md has example JSON; see Configuration for environment variables and Tutorials, Getting started for client wiring. Step-by-step: Cursor, Claude Desktop.
Optional components¶
Local diagram servers¶
For better performance or offline use:
PlantUML server¶
Kroki server¶
Then point UML-MCP at the local instances:
export USE_LOCAL_PLANTUML=true
export PLANTUML_SERVER=http://localhost:8080
export USE_LOCAL_KROKI=true
export KROKI_SERVER=http://localhost:8000
(On Windows, use set in cmd or $env:VAR = "value" in PowerShell instead of export.)
Troubleshooting¶
- Ensure Python 3.12 is installed and on your PATH
- Confirm dependencies installed (
uv syncor equivalent) - Verify any local servers are running
- Ensure write permissions if you use
output_dirwithgenerate_uml