AI coding agents

Let your coding agent build the backend.

Every SupaHost project comes with a hosted MCP server. Paste one config into Claude Code, Cursor, VS Code, Codex or Windsurf and your agent can design tables, write migrations, check security advisors and generate types — within the limits you set.

One-click setup

Copy a ready-made config for your client, or use the Cursor install link. Tokens stay in an environment variable, never in your repo.

Scoped access tokens

Give each agent its own token, limited to one project and the features it needs. Revoke it in one click.

Read-only mode

Start agents in read-only mode, enforced by Postgres itself, so they can inspect schemas and data without changing anything. Switch on write access when you're ready.

Tools that matter

List tables and extensions, run SQL, apply migrations, deploy Edge Functions and read their logs, run security and performance advisors, generate TypeScript types, inspect Storage and fetch project keys.

Token-level control

Each token records when it was last used. Revoke it in one click and the agent loses access immediately.

Works with SSH too

Prefer the terminal? Business instances add restricted SSH: psql into any project, read logs, and tunnel Postgres to your own tools.

How it works

  1. Create a token

    In the dashboard, create an access token for the project and choose read-only or read-write.

  2. Pick your client

    Choose Claude Code, Cursor, VS Code, Codex or Windsurf and copy the generated config.

  3. Start building

    Ask your agent to create tables, add RLS policies and generate types for your app.

Availability

All plans. SSH access is included on Business and Enterprise.

Compare plans

Questions

Which AI tools are supported?

Any client that speaks the Model Context Protocol over HTTP, including Claude Code, Cursor, VS Code (Copilot), OpenAI Codex and Windsurf. The dashboard generates the exact config for each one.

Can an agent damage my production database?

Only if you let it. Tokens can be read-only, limited to one project and limited to specific features. We recommend read-only tokens for production and read-write for development projects.

Can I see what an agent did?

Migrations applied through MCP are recorded with their names, and each token shows when it was last used. Revoke a token at any time.

Put every project on one plan.

Every plan starts with a 7-day free trial.