Open source · the backend your agents can use

The backend your AI actually understands.

Tea gives your AI application a real Postgres database with a REST API, MCP and permissions already wired up — plus a beautiful grid your team can edit by hand.

Start building free View on GitHub

Postgres native · MCP out of the box · Self-host in minutes

A Tea table of purchase requests, with a status column, assigned requesters, and AI-written summary and risk columns alongside the human ones.

Tea MCP

4 purchase requests summarised and scored.

Permissions verified

One backend, three ways in

One backend. Every client.

The UI, the REST API and the MCP tools are thin adapters over the same server functions. One implementation, one permission check, no drift between what a human sees and what an agent gets.

For your agents

MCP, already connected

Point your AI client at /mcp with a bearer token and it is connected — no OAuth dance. Catalogue-driven tools discover every table you add, now and in the future.

For your product

A REST API by default

Every table is instantly live under /api/v1. Filter, sort, search, paginate, import and export, with no endpoint to write.

For your team

A grid to see it all

Your backend comes with an admin UI. Edit text, dates, selects, ratings, people, files and relations right where they live — no migration to wait on.

Safe to hand to an agent

The right access, whoever is asking.

An agent is only as safe as the credential it carries. In Tea a token can never reach further than the person who minted it, so handing one to a model is not a leap of faith.

Precise by design

Manage, read and write are independent flags, not a ladder. An agent that writes rows need not be able to read them.

Down to the row

Scope a role to every row or only the ones it created. One agent's records stay invisible to the next.

Identity travels

A token does exactly what its owner can do and absolutely nothing they cannot — over REST, over MCP, in the UI.

Tea's role editor: the Approver role, with manage, read and write ticked independently per table and a Rows column scoping each grant.

Turn any table into a beautiful form.

Collect surveys, requests and sign-ups from people with no account — and no credential for your agents to handle. Responses land in the table your application already queries.

A Tea form titled Product feedback, asking what to build next and how important it is, with a rating and a Send feedback button.

Yours to run

Open source.
Postgres native.

Tea runs as one container against your own Postgres. It applies its own migrations safely and stores files in any S3-compatible bucket — so your application's data, and everything your agents touch, stays on your infrastructure.

Single containerAutomatic migrationsS3 compatible

Quick start

01$ docker compose up -d02$ pnpm install03$ pnpm db:push04$ pnpm dev

Ready at localhost:3000

Ship the backend

Start with one table.
Give your agent a key.

Create an account, model your data, and point your first agent at it in minutes.

Start building free