pg-aiguide 🔌 MCP server Open source

What it does

pg-aiguide gives AI coding tools deep, version-aware PostgreSQL knowledge so they write better database code. AI assistants often produce Postgres that is outdated, missing constraints and indexes, or unaware of modern features; pg-aiguide counters that by offering semantic search across the official PostgreSQL manual, curated best-practice "skills" that agents apply automatically, and extension ecosystem docs starting with TimescaleDB and PostGIS. It can be used as Agent Skills, as a public hosted MCP server, or as a Claude Code plugin.

Tools it gives your agent

The MCP server exposes search_docs, a unified search over multiple documentation sources (the PostgreSQL manual scoped by version, Tiger Data's TimescaleDB docs, and PostGIS), supporting both semantic vector similarity and BM25 keyword search. view_skill surfaces the curated, opinionated best-practice skills covering schema design, indexing strategies, and modern Postgres patterns, which agents draw on when generating or reviewing SQL. The skills can also be installed directly into an agent without the server, and the two combine for the deepest coverage.

Works with

Any MCP-aware coding agent, since the server is public and hosted. The README documents Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Windsurf, VS Code, Visual Studio, Goose, and LM Studio, and the skills install path lists more than forty agents.

How to install

claude plugin marketplace add timescale/pg-aiguide
claude plugin install pg@aiguide

Maintenance and safety

Apache-2.0 licensed, maintained by Timescale (Tiger Data), and updated recently. The MCP server is read-only documentation search hosted at mcp.tigerdata.com, so queries go to that endpoint; it does not touch or execute against your database. The skills are local guidance files.

Who should use it

Developers who want their coding agent to generate robust, well-indexed, version-appropriate PostgreSQL and TimescaleDB schemas and queries. It is most useful on projects where schema quality, correct indexing, and modern Postgres idioms matter, and where you would rather the agent consult authoritative documentation than guess from stale training data.