Supabase MCP Server

04 Apr 2025 · 9 minute read

By Greg Richardson, Engineering

What is an MCP Server?

MCP stands for Model Context Protocol. It standardizes how Large Language Models (LLMs) talk to platforms like Supabase.

An official Supabase MCP server has been launched. You can use this server to connect your favorite AI tools (such as Cursor or Claude) directly with Supabase.

Our MCP server connects your AI tools to Supabase so that they can perform tasks like launching databases, managing tables, fetching config, and querying data on your behalf.

Tools

MCP servers use Tools, which are a bit like "abilities". There are over 20 tools available in the Supabase MCP server.

You can:

• Design tables and track them using migrations
• Fetch data and run reports using SQL queries
• Create database branches for development (experimental)
• Fetch project configuration
• Spin up new Supabase projects
• Pause and restore projects
• Retrieve logs to debug issues
• Generate TypeScript types based on your database schema

For a full list of abilities, see Tools in the project README.

Setup

Configuration example for .cursor/mcp.json:

{  
  "mcpServers": {  
    "supabase": {  
      "command": "npx",  
      "args": [  
        "-y",  
        "@supabase/mcp-server-supabase@latest",  
        "--access-token",  
        "<personal-access-token>"  
      ]  
    }  
  }  
}

You'll need to create a personal access token (PAT) for the <personal-access-token> field. This token authenticates the MCP server with your Supabase account.

Some clients expect a slightly modified JSON format, and Windows users will have to prefix this command with cmd /c. For detailed step-by-step instructions for each client and OS, see the MCP documentation.

How does MCP work?

Most large language models (LLMs) today support "tool calling" where the model can choose to invoke a developer-provided tool based on the context of the conversation. MCP standardizes the tool ecosystem by creating a protocol understood by both clients (eg. Cursor) and tool providers (eg. Supabase), while decoupling them from each other.

Resources and prompts

MCP also incorporates optional primitives beyond tool calling: resources and prompts. Resources allow servers to expose arbitrary data and content that can be read by clients and used as context for LLMs. Prompts allow servers to define reusable prompt templates that clients can surface to users and LLMs.

What's next?

Create and deploy Edge Functions

Supabase Edge Functions allow you to run custom, server side code from the edge network. The ability to create and deploy Edge Functions directly from your favorite AI assistant is on the roadmap.

Native authorization

The latest revision of the MCP spec now includes official authorization support. Future versions will allow you to authenticate with Supabase using a standard OAuth 2 login flow, eliminating the need to manually create a personal access token.

Better schema discovery

Currently the MCP server provides a single list_tables tool. Future improvements will provide better structured access to other database objects like views, triggers, functions, and policies.

More protections

Additional protections like auto-detecting destructive operations and requiring confirmation before executing them are planned, leveraging Supabase's database branching capability.

---

Check out the docs for the latest updates and examples. For issues or feature requests, open an issue on the GitHub repo.