Supabase MCP Server

What it does

• Postgres tools: execute_sql with read-only / write / admin scopes, list_tables, extensions inspection
• Auth tools: list_users, get_user, create_user, update_user, delete_user (admin key required)
• Edge Function tools: deploy_function, list_functions, get_logs
• Storage tools: list_buckets, upload_object, download_object, copy_object, move_object
• Project-level access control via the Supabase personal access token (PAT)

When to use it

Best for end-to-end Supabase workflows — schema migrations under Claude's review, seeding test data, debugging RLS policies with read-only SQL, deploying edge functions from local code, and ad-hoc analytics on the project's own Postgres. Pairs with any agent framework that speaks MCP.

Getting started

npx -y @supabase/mcp-server-supabase@latest --project-ref=<your-project-ref>

Authenticate via the supabase login flow on first run; a personal access token is stored in the OS keychain. Add the entry to .mcp.json or ~/.claude.json.

Things to know

execute_sql with write or admin scope can drop tables — restrict the PAT to the minimum needed and use a separate project for development. The Edge Function deploy tool requires the function code to be present locally; it does not synthesize code from a prompt.