Home / MCP servers / Postgres
Hosted MCP server

Postgres MCP Server

Give your AI the power to query and update any PostgreSQL-compatible database, through one hosted MCP server you do not have to install or run. UniversalBench is the server: connect one URL and your AI gets Postgres plus every other capability, with safety built in. Works with Claude, ChatGPT, Gemini, and any MCP-compatible AI.

You do not have to host a Postgres MCP server

There are open source Postgres MCP servers you can download, configure, and run yourself, one capability at a time. UniversalBench takes that off your plate. It is a hosted MCP server, so there is nothing to install, nothing to keep running, and the same single connection also covers databases, code, web search, GitHub, files, and more.

Self hosted serverUniversalBench
SetupDownload, configure, run itPaste one URL
MaintenanceYou keep it runningHosted for you
CoveragePostgres onlyPostgres plus 20+ capabilities
SafetyYou secure itGuards built in
CredentialsWired into configEncrypted vault, never shown

Set up the Postgres MCP server in three steps

Connect the one MCP URL

Add the single UniversalBench MCP URL to your AI: Claude, ChatGPT, Gemini, or any MCP-compatible client. That one server carries this capability and every other.

Save your credential once

Take your postgresql:// connection string (any PostgreSQL-compatible database), then save it in your secret vault as DATABASE_URL. It unlocks the matching capability automatically and is never shown again.

Ask in plain language

Ask for what you want. The work runs on the hosted server and only the result returns to your chat.

# one hosted MCP server for Claude, ChatGPT, or Gemini https://universalbench-mcp.penantiaglobal.workers.dev/u/<your-key> # save your credential once in the secret vault DATABASE_URL = postgresql://user:password@host:5432/dbname
UniversalBench secret vault: secret name on the left, secret value on the right, then save
Any credential goes in your secret vault once: secret name on the left, secret value on the right, then Save.

See the Postgres MCP server in action

A real example. You ask a data question in plain language, your AI runs it on the live database, and only the rows that matter come back.

You Find the 20 customers with the highest refunds in the last 90 days, then tag them as review in the notes column. Your AI, through UniversalBench query select id, name, sum(refund) ... where created > now() - interval '90 days' group by id, name order by 3 desc limit 20 → 20 rows returned, not the whole table update notes = 'review' on those 20 ids done

You see the rows first, then change exactly those. Because the query runs on the server and only the result returns, a large table never floods your chat. Writes happen only when you ask for them.

Common things people build with the Postgres MCP server

What your AI can do

Frequently asked questions

Is there an official Postgres MCP server?

There are open source Postgres MCP servers you can self host. UniversalBench is the hosted option: one URL gives your AI Postgres and many other capabilities without running a server yourself.

Do I need to install or host anything?

No. UniversalBench is the hosted MCP server. You paste one URL into your AI and save your credential once in the secret vault, and it works.

Does it work with ChatGPT and Gemini?

Yes. It connects over the standard MCP protocol, so it works with Claude, ChatGPT, Gemini, and any MCP-compatible AI.

More MCP servers and guides

Get your Postgres MCP server free

One URL, first 1,000 calls free, no credit card.

Get your API key →