pgEdge MCP Server for Postgres

What is pgEdge MCP Server for Postgres?

The pgEdge MCP Server for Postgres enables SQL queries against PostgreSQL databases through MCP‑compatible clients such as Claude Desktop. Its Natural Language Agent lets you describe queries in plain English, which the server translates into SQL.

How to use pgEdge MCP Server for Postgres?

Clone the repository, run make build, then configure your MCP client (e.g., Claude Desktop) with the path to the binary. Provide database connection details via environment variables (PGHOST, PGPORT, PGDATABASE, PGUSER, PGPASSWORD) or a .pgpass file. The server can also be run as a standalone HTTP server or deployed with Docker.

Key features of pgEdge MCP Server for Postgres

• Read‑only protection – all queries run in read‑only transactions
• Resources, tools, and prompts for database interaction
• Production Go CLI client with Anthropic prompt caching (90% cost reduction)
• HTTP/HTTPS mode with token authentication and TLS support
• Web‑based management interface built with React
• Docker support for containerized deployment and hot‑reload of auth files

Use cases of pgEdge MCP Server for Postgres

• Schema discovery – retrieve table lists and column details with natural language
• Data analysis – ask for top customers, slow shipping orders, or aggregation queries
• System monitoring – check current connections, dead tuples, and cache hit ratios
• Internal developer tools – safely query production‑like databases without write access

FAQ from pgEdge MCP Server for Postgres

What databases does this server work with?

It connects to any PostgreSQL database using standard PG* environment variables.

Is the server safe for production or public‑facing applications?

No. The code is pre‑release and must not be put into production without thorough testing. It provides LLMs read access to your entire schema and data, so it should only be used for internal tools, developer workflows, or trusted environments.

What transports and authentication are supported?

The server supports stdio (subprocess) and HTTP/HTTPS transports. HTTP mode supports token‑based authentication and TLS encryption.

What runtime dependencies are required?

A PostgreSQL database and a running MCP client (e.g., Claude Desktop, the Go CLI client, or a custom HTTP client). LLM API keys (Anthropic, OpenAI, or Ollama) are needed for the Natural Language Agent features.

Are there any known limitations?

All queries are executed in read‑only transactions – write operations are not allowed. The server is currently in pre‑release status and may not be suitable for all workloads.