for AI agents

An MCP server that knows every component in this registry.

Install once and your AI agent can browse 203 components, configure them with the right params, hand you a ready-to-paste embed URL, and plan a full API install — without leaving the conversation.

Install the server Read the full docs

10 tools, zero OAuth scopes needed to browse
Runs over stdio — no server, no port, no daemon
Ships from npm as notion-arena-mcp

What the agent can do

Every tool is read-only by default. An install plan never asks for a Notion token until you approve it, and the final API call prints the exact requests it is about to make.

list_components

Browse every component with category, tier and capability filters.

search_components

Keyword search across name, tagline, description, tags and features.

get_component

Full detail for one slug — params, features, use cases, capabilities, delivery channels.

get_widget_schema

JSON Schema of every widget parameter, ready to drive a form or a config prompt.

build_embed_url

Given a slug + config, returns the exact /embed URL and the matching download URL.

plan_install

Dry-run (no token) install plan: every Notion API request, payload, manual step and warning.

get_registry_item

Raw shadcn-compatible registry JSON, including inline file contents.

get_install_guide

Human-readable install instructions for Embed URL, HTML block or Native recipe.

get_categories

All eight categories with emoji, tint and current counts.

get_site_info

Site endpoints, version, component counts, rate limits.

Install in two lines

The package is notion-arena-mcp on npm. Run it with npx so clients always pull the latest version; pin a version if you want reproducible tool behavior.

Claude Code

Travels with the repo — anyone who clones it gets the same tools.

bash

claude mcp add --scope project notion-arena -- npx -y notion-arena-mcp

Claude Desktop

Drop this into ~/Library/Application Support/Claude/claude_desktop_config.json.

json

{
  "mcpServers": {
    "notion-arena": {
      "command": "npx",
      "args": ["-y", "notion-arena-mcp@0.2.2"],
      "env": {
        "NOTION_ARENA_SITE_URL": "https://notionarena.vercel.app"
      }
    }
  }
}

Cursor / Windsurf

Save to .cursor/mcp.json in the project root.

json

{
  "mcpServers": {
    "notion-arena": {
      "command": "npx",
      "args": ["-y", "notion-arena-mcp@0.2.2"],
      "env": {
        "NOTION_ARENA_SITE_URL": "https://notionarena.vercel.app"
      }
    }
  }
}

Self-hosted? Add the env flag -e NOTION_ARENA_SITE_URL=https://your-host.com to point the server at your own deployment of the site. It works offline against the public catalog, but URLs and download links will resolve to the host you choose.

Or just paste this prompt into your agent

If you don't want to edit config files by hand, copy the block below and paste it directly into Claude Code, Cursor, ChatGPT with MCP support, or any other agent. It installs the server, restarts the client, and explains the decision rules so the agent picks the right channel first try.

prompt

Add the Notion Arena MCP server to this project so I can browse, configure and install Notion components from our chat.

Run this from the project root:

  claude mcp add --scope project notion-arena -- npx -y notion-arena-mcp

Once the server is connected:
- Use list_components / search_components to find the right component.
- Use get_component to inspect params.
- For embed/HTML installs, hand me the built URL with build_embed_url.
- For native recipes, always run plan_install in "plan" mode first and show me the request list + manual steps before asking for a Notion token.
- Defaults: prefer the HTML block for templates I sell (offline, frozen) and the Embed URL for internal dashboards (auto-updates). Never promise a widget can read or write Notion data.

Want to know exactly what each tool returns, how to pin a version, or how to author recipes that the MCP server can install? Read the MCP documentation

Building agents, not just using them?

Pair the MCP server with the Skills file to teach your agent the honesty policy, the three-channel decision table, and the localStorage / sandbox limits — before it makes a mistake.

See the skill