Sanity MCP

A headless CMS is a hard target for an agent because the useful operations all depend on a schema the agent cannot see. Sanity's official MCP server solves that first: it exposes the project's schema, so a model writing a GROQ query or a document patch is working against the real content model rather than pattern-matching from examples.

What it can do:

• Query and read — execute GROQ against any dataset in the project.
• Write — patch documents, create drafts and publish, with schema-aware field validation.
• Schema — fetch the deployed schema and deploy an updated one.
• Releases and datasets — manage content releases and switch between datasets, which is how staged publishing works in Sanity.
• Projects — list and inspect the projects the authenticated account can reach.
• AI features — generate and transform images, and run embeddings-based semantic search over content.
• Documentation lookup — pull Sanity documentation into context so the agent stops guessing at GROQ syntax.

That combination makes some genuinely tedious work tractable: bulk content migrations expressed as a described transformation, an editorial audit across a whole dataset, or scaffolding a content model from a brief and deploying it.

Connection: hosted at https://mcp.sanity.io over HTTP — no local process. In Claude Code: claude mcp add Sanity -t http https://mcp.sanity.io --scope user. Cursor and VS Code take the same {"type": "http", "url": "https://mcp.sanity.io"} entry in mcp.json. The quickest route of all is npx sanity@latest mcp configure, which writes the config for whichever client it detects.

Authentication: OAuth by default. For headless use, create a token at sanity.io/manage or through the CLI's token commands and pass it in an Authorization header.

A caution worth stating: this server can write to and publish content. Point it at a development dataset before you point it at production, and use a token scoped to what you actually want the agent to be able to change.