sanity-io / context

create-agent-with-sanity-context does the actual wiring: installing the Studio plugin, creating the Sanity Context document, and connecting an agent framework (Next.js, SvelteKit, Express, or Python are the ones the README walks through) to the resulting MCP URL. dial-your-context is a separate interactive session for writing the Instructions field that scopes what the server's groq_query and schema_explorer tools can see — the fix path when an agent's answers are wrong or it seems confused about how your content's schema relates. shape-your-agent is a third interactive session, for the system prompt itself: tone, refusal behavior, and how the agent should communicate, independent of what data it can reach.

This is specifically for building an agent that serves your own end users — search, a support bot, a content-driven feature in your app — not for using an AI coding assistant against your Sanity workspace; the README is explicit that its other MCP endpoint, the Sanity MCP Server, is the one for that. Getting any of the three doing something real needs a Sanity project with a deployed Studio (v5.1.0 or newer), a Sanity API read token, and your own LLM API key — none of which these three skills provide; they assume the account and the content already exist and help you wire and tune the agent around them.

READ THE FULL ANALYSIS＋

Two of the three need a live connection; the material doesn't say which one doesn't. Our dependency records flag two of the three skills as needing the Sanity Context MCP server actually running, and one as needing nothing beyond the skill file itself — but neither the skill descriptions nor the README identify which is the exception, so we aren't guessing.

Checked 19 September 2026 from the repository README and the 3 skills currently live in our database; we found no license recorded for this repository.