Postgres for AI agents
Agents create, query and delete Postgres databases through the API, CLI or MCP server, with scoped keys and scale to zero built in.
Postgres your agents create for themselves
The database primitives agents need: a Postgres per task, created through the API, CLI or MCP server, that costs no compute when nobody is using it.
Ready for AI-assisted coding
Claude Code, Cursor, Codex and other coding agents work with Xata through the MCP server and CLI.
Quick to start
Hand setup to your coding agent. It installs the CLI, creates a project and connects your app.
Building a platform that gives your users Postgres?
White-label Postgres with branching and scale to zero. See Platforms.
Add a database to your agent
Provisioning is one call
An agent that needs a scratch database should not wait on a ticket or share one with every other run. With Xata it creates one when the task starts and deletes it when the task ends. A new database is ready in about a second. A branch with a copy of real data takes about three seconds.
$ xata branch create --name my-test --no-parent
✓ Postgres ready · ~1s
postgresql://xata@db.xata.tech/my-test
280 branches
Active: 280Idle: 0
Compute$26.88/hr
Many short-lived databases, no idle bill
Agent workloads look different from apps. Instead of one database that lives for years, you get many that live for minutes. Fast provisioning makes that practical, and scale to zero keeps the idle ones from adding up.
Idle databases cost nothing
Turn on scale to zero and an idle database releases its compute. The next connection wakes it.
Code with AI
Your coding agent can manage the database too
Agents already write the queries. With Xata they can also create the branch to run them on.
In your editor
Works with the agents you already use
Create a branch, run a migration, read the schema and throw the branch away, all from Claude Code, Cursor, Codex, VS Code (Copilot), Windsurf, Cline, OpenCode or Zed.
Agent guides →MCP serverConnect MCP →
One hosted endpoint for every client
Connect with OAuth or an API key. Read, write and destructive calls are separate tools, and SQL is read-only by default.
Branching workflows for coding agents
Get started
Let your agent set it up
Paste one prompt into your coding agent and it walks you through the setup, asking before it creates anything billable.
01
Install the CLI
Your agent asks first, then installs it.
02
Log in
Sign in through the browser, or use a scoped API key in CI.
03
Create a project and branch
The agent proposes the settings and asks before creating anything billable.
04
Connect your app
Credentials go into a gitignored env file, never into the chat.
Give every agent task its own Postgres
Create a database when the task starts, delete it when it ends.