Remote and Docker deployment¶
By default each user's AI client starts its own copy of the server (the stdio transport). To share one server, for
example for a team or a web client, run it over HTTP.
Run over HTTP¶
uvx mcp-trino-python --transport http --host 0.0.0.0 --port 8000
- MCP endpoint:
http://<host>:8000/mcp(Streamable HTTP) - Health check:
http://<host>:8000/health
Connect a client to it, for example in Claude Code:
claude mcp add --transport http trino http://localhost:8000/mcp
There is no authentication on the HTTP endpoint
Anyone who can reach the port can run queries as TRINO_USER. Keep it on a private network or put it behind a
proxy that authenticates users, and consider read-only mode.
Docker¶
The image is ghcr.io/alaturqua/mcp-trino-python. Tags follow the releases (0.8.1, 0.8, latest).
# Streamable HTTP on port 8000
docker run --rm -p 8000:8000 \
-e TRINO_HOST=trino.example.com -e TRINO_USER=analyst \
ghcr.io/alaturqua/mcp-trino-python --transport http
Without arguments the container uses stdio, which is what MCP clients expect when they start it themselves.
Docker Compose¶
The repository's docker-compose.yml starts Trino with sample data plus the server on port 8000:
git clone https://github.com/alaturqua/mcp-trino-python.git
cd mcp-trino-python
docker compose up
Then connect your client to http://localhost:8000/mcp.
Legacy SSE¶
Some older clients only support the SSE transport. Use --transport sse; the endpoint is then /sse.
New setups should use http.