Supabase MCP Server
Give your AI agents a Postgres backend — query, manage, and build on Supabase.
Updated
What is Supabase MCP?
The Supabase MCP server is a Model Context Protocol server that lets AI agents and coding assistants work directly with your Supabase projects. With it connected, an agent can run SQL queries, create and modify tables, manage database branches, read logs, and generate TypeScript types — turning Supabase into a tool the model can act on rather than something you copy schema out of.
It runs locally via npx (stdio) and authenticates with a Supabase personal access token. You can scope it to a single project and run it in read-only mode for safety.
Tools the Supabase MCP server exposes
| Tool | What it does |
|---|---|
| list_tables | List tables, columns, and relationships in the database |
| execute_sql | Run a SQL query against the project's Postgres database |
| apply_migration | Apply a DDL migration to the database |
| list_branches | List development branches for the project |
| create_branch | Create a new development branch |
| get_logs | Fetch logs for the project (API, Postgres, auth, etc.) |
| generate_typescript_types | Generate TypeScript types from the current schema |
Connect the Supabase MCP server
Connect Supabase MCP to Cursor
- 1
Create a token
In the Supabase dashboard, generate a personal access token under Account → Access Tokens.
- 2
Add to .cursor/mcp.json
Add a `supabase` server running `npx -y @supabase/mcp-server-supabase@latest --project-ref=<ref> --read-only`, passing the token via the `SUPABASE_ACCESS_TOKEN` env var.
- 3
Reload
Reload Cursor's MCP settings and confirm the Supabase tools appear.
Start with `--read-only` and a single `--project-ref` so the agent can't touch production by accident.
Connect Supabase MCP to Claude Code
- 1
Add the server
Run `claude mcp add supabase -- npx -y @supabase/mcp-server-supabase@latest --project-ref=<ref>` with `SUPABASE_ACCESS_TOKEN` set.
- 2
Confirm
Ask Claude to "list my Supabase tables" to verify the connection.
Sample use cases
You're scaffolding a feature in Cursor
The agent inspects your schema, writes the migration, applies it to a branch, and generates matching TypeScript types.
Debugging a production error
The agent pulls Postgres and API logs through the MCP and points to the failing query.
Ad-hoc analytics question
Ask in plain English; the agent writes and runs the SQL read-only and returns the answer.
Security & permissions
The Supabase MCP server authenticates with a personal access token that you supply as an environment variable — keep it out of source control. Two flags materially reduce risk: `--read-only` restricts the agent to read queries, and `--project-ref` pins it to a single project so it can't reach the rest of your account. Supabase recommends connecting to a development project, not production, when using MCP with autonomous tools. Rotate the token from the dashboard if it leaks.
Using Supabase MCP with Gamut
On Gamut, the Supabase MCP server lets agents use your database as part of an automated, event-driven workflow. An agent can be triggered by a webhook, query Supabase for the relevant rows, enrich or transform the data using other connected MCP servers, and write results back — on a schedule or in response to events, with the connection and credentials managed by Gamut rather than living in a local config file.
Frequently asked questions
What is the Supabase MCP server?
An open-source Model Context Protocol server that connects AI agents to your Supabase projects so they can run SQL, manage schema and branches, read logs, and generate types.
Is the Supabase MCP server free?
Yes — it's free and open source. You authenticate with a Supabase personal access token and normal project limits apply.
How do I connect Supabase MCP to Cursor?
Add a `supabase` MCP server in `.cursor/mcp.json` that runs `npx @supabase/mcp-server-supabase` with your access token and `--project-ref`, ideally with `--read-only`.
Is the Supabase MCP server safe for production?
Use `--read-only` and pin `--project-ref` to a development project. Avoid pointing autonomous agents at production data.