Announcing the Grafbase remote MCP server

Tom HouléTom Houlé
Announcing the Grafbase remote MCP server

Back in April, we were the first GraphQL platform to release Model Context Protocol support. We've followed the developments in the MCP world, and extended support to the latest version of the spec, released in June. We are now also exposing an MCP server for the Grafbase API that powers the Grafbase Dashboard and the CLI.

This MCP server is available at https://api.grafbase.com/mcp. Here's how to configure it in Cursor, to take one popular example:

{ "mcpServers": { "grafbase": { "url": "https://api.grafbase.com/mcp" } } }

or with Claude Code:

claude mcp add grafbase https://api.grafbase.com/mcp --transport=http

Then log in with /mcp inside Claude Code.

You'll notice that we only specify a URL, no access token. This is because the Grafbase remote MCP implements the OAuth 2.1 based MCP authorization flow, where your MCP client will open a browser window to authenticate with your Grafbase account when it needs to.

Here's a few example questions you can ask:

  • Make a table with my graphs on Grafbase and how much traffic their production branches have had in the last 7 days.
  • Is there a ProductReviews type defined in the api@next schema?
  • What subgraphs have been published in the last 7 days on my my-project@production graph?

You can even ask for mutations:

  • Please delete the "experiment" branch on the "api" graph for me.
  • Add "jane@example.com" to the "backend" team.
  • Run a schema check for api@main with the schema.graphql file from this repo.

This new remote MCP server is the MCP server that is part of the Grafbase Gateway, in front of the Grafbase GraphQL API.

All the components used by the Grafbase remote MCP are available to you, as part of the Gateway, or open source extensions. Check out our MCP docs page for more details.

We are really excited about the possibilities opened by MCP, and we will continue working on making your GraphQL API ready to power rich, secure and fast interactions with MCP client users and agents.

If you have opinions, use cases, feedback, or just want to chat about MCP and AI with us, we are always excited to have these conversations. So please reach out!