sanity-io / agent-toolkit

Collection of resources to help AI agents build better with Sanity (opens in a new tab). Supports Cursor, Claude Code, Codex, VS Code, Lovable, v0, Replit, OpenCode, and any other editor/agent compatible with MCP or Agent Skills (opens in a new tab).

Features

• MCP server: Direct access to your Sanity projects (content, datasets, releases, schemas) and agent rules.
• Agent skills: Comprehensive best practices skills for Sanity development, content modeling, SEO/AEO, and experimentation. Includes 21 integration/topic guides and 26 focused best-practice rules.
• Agent Plugin: plugin.json, skills/, and mcp.json components for any Agent Plugins-compatible client.
• Claude Code plugin: MCP server, agent skills, and slash commands for Claude Code users. Available on the official Anthropic plugin marketplace (opens in a new tab).
• Cursor plugin: MCP server, agent skills, and commands on the Cursor Marketplace (opens in a new tab).
• Codex plugin: MCP server and agent skills for OpenAI Codex (opens in a new tab) users.

Get started

Choose your path based on how you want agents to work with Sanity:

1. MCP server — Give your agent always up-to-date rules and full access to your Sanity projects. No local files to maintain. Works with Cursor, VS Code, Claude Code, Lovable, v0, Replit, OpenCode, and other MCP-compatible clients.
2. Agent skills — Install best practices skills for Sanity, content modeling, SEO/AEO, and experimentation. Works with Cursor, Claude Code, and any Agent Skills (opens in a new tab)-compatible agent.
3. Plugin — Install the Sanity plugin for Cursor, Claude Code, or any Agent Plugins-compatible client. Bundles MCP server, agent skills, and commands.
4. Manual installation — Copy the skill references locally for offline use. You’ll need to update them yourself.

Option 1: Install MCP server (recommended)

Give agents direct access to Sanity projects and always up-to-date agent rules via the MCP server.

Quick install via Sanity CLI

Run in terminal to detect and configure MCP for Cursor, Claude Code and VS Code automatically:

npx sanity@latest mcp configure

Uses your logged-in CLI user for authentication — no manual tokens or OAuth needed.

Client-specific instructions

One-click install: (opens in a new tab)

Or manually: Open Command Palette (Cmd+Shift+P / Ctrl+Shift+P) → View: Open MCP Settings → + New MCP Server → add to mcp.json:

{
  "mcpServers": {
    "Sanity": {
      "type": "http",
      "url": "https://mcp.sanity.io"
    }
  }
}

Run in terminal. Authenticate with OAuth on next launch:

claude mcp add Sanity -t http https://mcp.sanity.io --scope user

Run in terminal. Authenticate with OAuth on next launch:

codex mcp add Sanity --url https://mcp.sanity.io

Or manually add to ~/.codex/config.toml:

[mcp_servers.Sanity]
url = "https://mcp.sanity.io"

Open Command Palette (Cmd+Shift+P / Ctrl+Shift+P) → MCP: Open User Configuration → add:

{
  "servers": {
    "Sanity": {
      "type": "http",
      "url": "https://mcp.sanity.io"
    }
  }
}

Sanity is available as a prebuilt chat connector in Lovable:

1. Open Connectors → Chat connectors
2. Select Sanity
3. Click Connect and sign in to authorize your Sanity account

In your next prompt, reference your Sanity project or ask the agent to read your schema.

See the Lovable MCP documentation (opens in a new tab) or Sanity + Lovable guide (opens in a new tab) for more details.

In the prompt input field, click Prompt Tools → MCPs → Add New → Select Sanity → Authorize → Authenticate with OAuth.

Go to Integrations Page (opens in a new tab) → scroll to MCP Servers for Replit Agent → Add MCP server → Enter Sanity as name and https://mcp.sanity.io as Server URL → Test & Save → Authenticate with OAuth.

Add to your opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "sanity": {
      "type": "remote",
      "url": "https://mcp.sanity.io",
      "oauth": {}
    }
  }
}

Then run: opencode mcp auth sanity

For any MCP-compatible client, add https://mcp.sanity.io as the server URL.

If your client doesn’t support remote MCP servers, use a proxy like mcp-remote:

{
  "mcpServers": {
    "Sanity": {
      "command": "npx",
      "args": ["mcp-remote", "https://mcp.sanity.io", "--transport", "http-only"]
    }
  }
}

Authorization and troubleshooting

Manual MCP configuration uses OAuth by default. You can use token auth instead by setting an Authorization: Bearer <token> header in the MCP config. If authentication fails after CLI setup, rerun npx sanity@latest mcp configure and restart your MCP client. For OAuth reset issues, Cursor provides Cursor: Clear All MCP Tokens and VS Code provides Authentication: Remove Dynamic Authentication Providers.

See the Sanity MCP docs (opens in a new tab) for authorization options and troubleshooting.

Option 2: Install Agent Skills

Install best practices skills that work with any Agent Skills (opens in a new tab)-compatible agent.

npx skills add sanity-io/agent-toolkit

See Option 3 for plugin installation.

Option 3: Install plugin

Install the Sanity plugin to get MCP server, agent skills, and commands. Available on the Claude Code marketplace (opens in a new tab), Cursor Marketplace (opens in a new tab), or from this repo as an Agent Plugin (opens in a new tab).

Claude Code

The Sanity plugin is listed on the official Anthropic plugin marketplace (opens in a new tab). The official marketplace (claude-plugins-official) is pre-registered when you start Claude Code — you do not need to add a custom marketplace.

Install from Claude Code:

/plugin install sanity@claude-plugins-official

If the plugin is not found, refresh the marketplace catalog and retry:

/plugin marketplace update claude-plugins-official

Then run /reload-plugins to activate without restarting.

Alternative: interactive install

1. Run /plugin and open the Discover tab
2. Search for Sanity
3. Review what the plugin will install — commands, skills, hooks, and MCP servers — before confirming (Anthropic recommends reviewing plugin permissions and source before installing (opens in a new tab))
4. Choose an installation scope:
   • User (default): all projects on this machine
   • Project: shared with collaborators via .claude/settings.json
   • Local: this repository only
5. Run /reload-plugins to activate without restarting

Verify installation: Ask Claude Code: “which skills do you have access to?”

You should see the Sanity skills listed.

Start using: Use natural language and skills activate automatically:

Help me create a blog post schema in Sanity

Review my GROQ query and Next.js Visual Editing setup

Or run /sanity to explore all capabilities.

Cursor

Install from the Cursor Marketplace (opens in a new tab) by running this in Cursor chat:

/add-plugin sanity

Verify installation: Ask Cursor: “which skills do you have access to?”

You should see the Sanity skills listed.

Start using: Use natural language and skills activate automatically:

Help me create a blog post schema in Sanity

Review my GROQ query and Next.js Visual Editing setup

Codex

1. Add the Sanity marketplace:

codex plugin marketplace add sanity-io/agent-toolkit

2. Install the plugin from Codex’s plugin directory (select the Sanity Agent Toolkit marketplace, then install Sanity).

3. Restart Codex. Verify by asking: “which skills do you have access to?” — you should see the Sanity skills listed.

Option 4: Manual installation

Install the skill references locally to teach your editor Sanity best practices:

1. Copy skills/sanity-best-practices/ to your project.
2. (Recommended) Copy AGENTS.md to your project root to act as a knowledge router.

Capabilities

MCP tools

With MCP connected, your AI can use tools like:

• query_documents — run GROQ queries directly
• create_documents — create draft documents from structured content, or version documents when a release ID is provided
• patch_documents — surgical edits to existing documents; published documents are edited by creating or updating drafts
• publish_documents / unpublish_documents — manage document lifecycle
• deploy_schema / get_schema — deploy MCP-managed schemas and inspect deployed schemas
• deploy_studio — deploy a hosted Studio bound to an MCP-managed schema
• create_release / list_releases — create and inspect Content Releases
• create_version — create version documents for releases
• generate_image / transform_image — AI image generation and editing
• whoami — verify the authenticated Sanity user
• get_project_studios — list Studio applications linked to a project
• search_docs / read_docs — search and read Sanity documentation
• list_sanity_rules / get_sanity_rules — load agent rules on demand
• give_feedback — report MCP tool errors, missing capabilities, confusing output, or documentation issues

MCP-managed schemas are resolved before Studio-deployed and legacy schemas. If you deploy schema changes with deploy_schema, redeploy any matching MCP-managed Studio with deploy_studio so it picks up the latest schema. generate_image, transform_image, and create_version with an instruction consume Sanity AI credits.

See the full list of available tools (opens in a new tab).

Agent skills

Best practices skills that agents like Claude Code, Cursor, GitHub Copilot, etc. can discover and use automatically. Skills follow the Agent Skills (opens in a new tab) format. See Option 2 for installation.

Getting started flow

The onboarding guide follows three phases:

1. Studio & Schema — Set up Sanity Studio and define your content model
2. Content — Import existing content or generate placeholder content via MCP
3. Frontend — Integrate with your application (framework-specific)

Just say: “Get started with Sanity” to begin.

Slash commands (Claude Code)

Repository structure

Note: The reference files in skills/sanity-best-practices/references/ are the canonical content for the Sanity MCP server’s list_sanity_rules / get_sanity_rules tools. Each file must have valid name and description frontmatter — rule names are derived from filenames (e.g., nextjs.md → nextjs).

sanity-io/agent-toolkit/
├── AGENTS.md                      # Knowledge router & agent behavior
├── README.md                      # This file
├── plugin.json                    # Portable Agent Plugins v1 manifest
├── mcp.json                       # Portable Agent Plugins v1 MCP configuration

All skills use references/ for detailed content loaded on demand. The sanity-best-practices references are also the canonical source for the MCP server’s Sanity rules.

Resources

• Create Sanity account (opens in a new tab)
• Sanity documentation (opens in a new tab)
• GROQ language reference (opens in a new tab)
• Visual Editing guide (opens in a new tab)
• Sanity TypeGen (opens in a new tab)
• MCP server docs

Contributing

Found a better pattern? Missing a framework or best practice? Read the contributing guide (opens in a new tab) for how skills work and what makes a good contribution, then:

1. Fork the repo.
2. Install dependencies with npm install.
3. Make your changes in skills/<skill-name>/.
4. Run npm run validate:all to check skill and plugin validity.
5. Submit a PR.

Support

• Sanity Community (Discord) (opens in a new tab)
• GitHub issues (opens in a new tab)

License: MIT