Agent Skills

dial-your-context

Tools for building AI agents with structured access to Sanity content

安装方式
CLI
npx skills add https://github.com/sanity-io/agent-context --skill dial-your-context

使用 CLI 安装这个技能，并在你的工作区中直接复用对应的 SKILL.md 工作流。

最后更新于 4/22/2026

Sanity Agent Context

Give AI agents structured access to your content. Agent Context is a hosted MCP endpoint that connects AI agents to your Sanity Content Lake, where content is stored as structured, queryable data (not pages or blobs of HTML).

Instead of vectorizing your content into embeddings and hoping similarity search returns the right answer, Agent Context lets agents query your actual data model: filter by fields, traverse references between documents, and combine structured queries with semantic search. Embeddings for exploration, structured queries for precision.

Read the full documentation →

Agent Context vs Sanity MCP Server — Sanity offers two MCP endpoints. The Sanity MCP Server gives AI coding assistants like Cursor, Claude Code, and v0 full access to your Sanity workspace (content, schemas, releases, and more). Agent Context is different: it's for production agents that serve your end users — read-only, scoped access you use to power search, support bots, and other content-driven features in your application.

How it works

flowchart LR
  A["Your agent"] <-->|"MCP"| B["Agent Context <br> (hosted by Sanity)"]
  B --> C["Your content in Sanity"]

You create an Agent Context document in Sanity Studio. This document controls what content your agent can access and generates a unique MCP URL. Your agent connects to that URL with an API token.

Agent Context exposes three MCP tools:

Tool What it does
initial_context Returns a compressed schema overview: content types, fields, and document counts
groq_query Runs GROQ queries with optional semantic search
schema_explorer Returns the full schema for a specific content type

With these tools, your agent can:

  • Look up exact prices, inventory, or metadata (not approximate text matches)
  • Filter products by category, size, color, or any field in your schema
  • Follow references between documents (a product's brand, a brand's products)
  • Combine structured filters with semantic search ("trail running shoes under $150")

Here's a combined query in GROQ:

*[_type == "product" && category == "shoes"]
  | score(text::semanticSimilarity("lightweight trail runner for rocky terrain"))
  | order(_score desc)
  { _id, title, price, category }[0...5]

Structural filter (category == "shoes") for precision. Semantic ranking (text::semanticSimilarity()) for discovery.

Get started

Prerequisites

  • A Sanity project with content and a deployed Studio (v5.1.0+)
  • A Sanity API read token — create one at sanity.io/manage (Project → API → Tokens) or via CLI:
    npx sanity tokens add "Agent Context" --role=viewer
  • An LLM API key (Anthropic, OpenAI, or another provider)

New to Sanity? Start here.

Using skills

If you're using Claude Code, Cursor, or similar, you can install skills that guide your AI assistant through the setup:

npx skills add sanity-io/agent-context --all

Then prompt:

Use the create-agent-with-sanity-context skill to help me build an agent.

The skill walks you through Studio setup, MCP connection, and configuration for your stack (Next.js, SvelteKit, Express, Python, etc).

Other skills help you refine: dial-your-context (tune the Instructions field) and shape-your-agent (craft a system prompt).

Manual setup

  1. Install the Studio plugin:

    npm install @sanity/agent-context

    // sanity.config.ts
import {defineConfig} from 'sanity'
import {agentContextPlugin} from '@sanity/agent-context/studio'

export default defineConfig({
  // ...existing config
  plugins: [agentContextPlugin()],
})

  2. Create an Agent Context document in Studio and copy the MCP URL.

  3. Connect your agent using any MCP-compatible framework. Example with Vercel AI SDK:

    import {createMCPClient} from '@ai-sdk/mcp'

const mcpClient = await createMCPClient({
  transport: {
    type: 'http',
    url: process.env.SANITY_CONTEXT_MCP_URL,
    headers: {
      Authorization: `Bearer ${process.env.SANITY_API_READ_TOKEN}`,
    },
  },
})

Troubleshooting

Validate the connection — Test that your token and endpoint work:

curl -X POST https://api.sanity.io/v2026-03-03/agent-context/:projectId/:dataset/:slug \
  -H "Authorization: Bearer $SANITY_API_READ_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc": "2.0", "method": "tools/list", "id": 1}'

If this returns a list of tools, you're connected. The full MCP URL is shown in your Agent Context document in Studio.

401 Unauthorized — Your SANITY_API_READ_TOKEN is missing or invalid. Generate a new token at sanity.io/manage → Project → API → Tokens.

No schema or empty results — Agent Context requires a deployed Studio. Run npx sanity deploy. If you've set a content filter, ensure it matches published documents.

Tools not appearing — Verify the MCP URL is correct (project ID, dataset, slug) and that the Agent Context document is published.

Learn more