A Postgres MCP server lets an AI model read, query and analyse any PostgreSQL database through the Model Context Protocol. Point it at your connection string and ask questions in plain English — the model writes the SQL, runs it, and explains the result.
Bring your own
Claude Sonnet 4.5
Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side and balance quality vs. cost.
Your Postgres connection string. Run the open-source crystaldba/postgres-mcp locally, or use a hosted option like Neon MCP.
How models use it and what it is built for.
A Postgres MCP server wraps a PostgreSQL connection and exposes its capabilities as MCP tools — listing tables, describing schemas, running parameterised SQL, explaining query plans, and (optionally) writing rows back. Models like Claude, GPT and Gemini can then plan multi-step questions ("which customers churned last quarter and why?") and execute them as a sequence of tool calls. The most popular implementations are crystaldba/postgres-mcp (open-source, MIT) and the hosted Neon MCP server, which lets you query Neon-hosted Postgres without running anything locally.
Typical tools an AI model can call. Exact names vary by version.
Copy any of these into MCP Agent Studio after connecting.
List all tables in the public schema and tell me which ones look like fact tables vs dimensions.
Show me the top 10 customers by total order value in the last 90 days.
There is a slow query on the orders table — can you find missing indexes?
Summarise the schema of the users table and flag any columns that look like PII.
This is not a single-model product: you get the same MCP connection with 60+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for PostgreSQL
Claude Sonnet 4.5
Claude Sonnet 4.5 handles multi-step SQL planning and schema reasoning reliably. For cheap, high-volume reads use Haiku 4.5 or GPT-5.2 mini.
Listing tools proves the server is reachable, not that a model can work with it. Evals go further: they read every tool on the server, write a test suite from its real schemas, and run it — code decides pass/fail on the responses (schema conformance, error codes, pagination, result caps) while a scoring model grades plain-English tasks driven through the tools.
Get a pass/fail report per tool with the evidence behind each verdict — and replay the same suite after every schema change. Destructive tools are excluded from the run.
Run evalsOpen MCP Agent Studio with the connection pre-filled. Add your token, pick any of 60+ models, and start chatting — no install required.
Try PostgreSQL in Agent StudioCommon questions about connecting, scoping and using it safely.
It is an MCP-compatible bridge between an AI model and a PostgreSQL database. The server exposes tools like list_tables, describe_table and execute_sql so models can answer database questions by writing and running SQL on your behalf.
There is no single canonical hosted server — Postgres connection strings are private. The most-used implementations are crystaldba/postgres-mcp (run locally with your DATABASE_URL) and Neon MCP (hosted, only works against Neon-managed databases).
Use a read-only role, restrict the server to a single schema or set of tables, and never paste service-role credentials. For exploration, point it at a staging copy. Most Postgres MCP servers also support a SQL allowlist or read-only mode.
Claude Sonnet 4.5 is the strongest at multi-step SQL reasoning. For cheaper, high-volume read queries Claude Haiku 4.5 and GPT-5.2 mini are usually sufficient. Compare them side-by-side in MCP Agent Studio before committing.
Open MCP Agent Studio, paste the URL of a running Postgres MCP server (HTTP or SSE), pick a model and start chatting. You can also bring up Neon MCP and connect to it without installing anything locally.