Supabase MCP Server enables LLMs to talk to external services via the Model Context Protocol (MCP). It connects AI assistants directly with a Supabase project, allowing them to perform tasks such as managing tables, fetching configuration, and querying data using a curated set of tools. Tools are grouped into feature groups (account, docs, database, debugging, development, edge functions, branching, storage) that can be enabled or disabled via the features query parameter. The server can be run in the cloud at mcp.supabase.com/mcp or locally with Supabase CLI at http://localhost:54321/mcp, and self-hosted deployments are supported. You can scope access to a specific project with project_ref to limit exposure to that project; without scoping, the server will have access to all projects in your account. Read-only mode can be enabled to restrict mutating actions and execute queries as a read-only Postgres user. The toolset includes operations such as listing and creating projects, listing tables, applying migrations, deploying Edge Functions, and managing branches. Security guidance covers prompt-injection risks and best practices like reviewing tool calls before execution. The MCP server is pre-1.0, so breaking changes may occur, and a separate PostgREST MCP server exists for different use cases.