fluid-postgres-mcp

# fluid-postgres-mcp [](LICENSE) [](https://www.python.org/downloads/) A PostgreSQL [MCP](https://modelcontextprotocol.io/) server for AI agents. Streams large result sets to CSV, enforces per-query timeouts, auto-reconnects with backoff, and supports long-running tunnel scripts (e.g. AWS SSM port-forwarding) with credential rotation. Fork of [crystaldba/postgres-mcp](https://github.com/crystaldba/postgres-mcp). ## Install > Jump to: [Other AI agents](#other-ai-agents) · > [Alternative install methods](#alternative-install-methods) · > [Develop](#develop) Python 3.10+. Published on PyPI as [`fluid-postgres-mcp`](https://pypi.org/project/fluid-postgres-mcp/); console entry point of the same name. ### Verify install Before wiring the MCP into your agent, confirm the install actually resolved its dependencies: ```bash uvx fluid-postgres-mcp --version # prints "fluid-postgres-mcp X.Y.Z", exit 0 uvx fluid-postgres-mcp --help # prints usage, exit 0 ``` Exit 0 from either command means the package downloaded and every runtime dependency imported successfully. A Python traceback or non-zero exit means at least one import failed — typically a missing system library (e.g. `libpq` on minimal Linux images) or a broken `uvx` cache. Fix that before continuing; an agent registration against a broken install fails silently at first tool call. ### With Claude Code (primary) ```bash claude mcp add fluid-postgres-mcp -- \ uvx fluid-postgres-mcp \ postgresql://reader:pw@host:5432/db ``` With a long-running tunnel script (see [Pre-connect scripts](#pre-connect-scripts) for the protocol the script must speak): ```bash claude mcp add fluid-postgres-mcp -- \ uvx fluid-postgres-mcp \ --pre-connect-script /path/to/your-tunnel.sh ``` ### Other AI agents Brief one-shot snippets — copy-paste, or read your agent…