Development¶
Setup¶
git clone https://github.com/alaturqua/mcp-trino-python.git
cd mcp-trino-python
uv sync
Tests¶
The tests run against a real Trino in Docker:
docker compose up -d --wait trino
uv run pytest
Without a reachable Trino, the server tests are skipped and only the SQL-generation tests run.
Lint and format¶
uv run ruff check
uv run ruff format
Try your changes¶
uv run mcp-trino --transport http # http://localhost:8000/mcp
npx @modelcontextprotocol/inspector # browse tools and call them by hand
Docs¶
uv run --group docs mkdocs serve # http://127.0.0.1:8000
The site is published to GitHub Pages on every push to main that touches the docs.
Project layout¶
| Path | What it is |
|---|---|
src/mcp_trino/server.py |
MCP tools, prompts and the command-line entry point |
src/mcp_trino/trino_client.py |
SQL generation and query execution |
src/mcp_trino/config.py |
Environment variable handling |
server.json |
MCP Registry listing |
.claude-plugin/ |
Claude Code plugin and marketplace |
mcpb/ |
Claude Desktop extension manifest and build script |
Releasing¶
- Bump
versioninpyproject.tomland.claude-plugin/plugin.json. - Publish a GitHub release tagged
v<version>.
The Publish workflow then pushes the package to PyPI, the image to ghcr.io, the listing to the MCP Registry, and
attaches the Claude Desktop extension to the release. Build the extension locally with sh mcpb/pack.sh
(needs npm install -g @anthropic-ai/mcpb).