Skip to content
Nerdy beta

MCP Server

This collection exposes a Model Context Protocol server so your AI tools can query it directly: searching notes, reading full content, browsing by tag or category, and following backlinks. Every note in the collection is public, so the read tools need no account. Point your client at the endpoint and start asking. The builder tools write to documents you own, so those ask for a free nerdnoir.ai account the first time you call one.

claude mcp add --transport http --scope user nerdy https://nerdnoir.ai/api/mcp

The Cursor button hands off to the app (if nothing happens, it is not installed). Claude Desktop and ChatGPT need a few manual steps; the setup guide covers both. Requires a free nerdnoir.ai account, and your client prompts you to sign in once.

Endpoint

https://nerdnoir.ai/api/mcp

Transport: streamable HTTP (MCP 2025-11-25 spec).

What the Server Exposes

Tools

ToolWhat it does
search_collectionKeyword search. Returns ranked hits with title, slug, category, tags, excerpt, and relevance score.
get_noteFetch the full markdown body of a note by slug.
list_by_tagAll notes matching a given tag.
list_by_categoryAll notes in a top-level category (concepts, tools, workshops, etc.).
list_backlinksNotes that wikilink to a given slug.

Resources

Every published note is addressable as collection://note/{slug}. Clients that support the resources primitive can browse the collection natively.

Connecting Claude Desktop

Claude Desktop supports custom connectors — remote MCP servers you add straight from settings. No local bridge, no npx, no config file to edit. If you use the builder tools, it also runs the OAuth sign-in for you.

  1. Open Claude Desktop and go to Settings → Connectors.
  2. Click + then Add custom connector.
  3. Enter a name (nerdy) and the server URL: https://nerdnoir.ai/api/mcp
  4. Click Add. If a sign-in prompt appears, complete it; you need it for the builder tools, not for reading.

The collection’s tools show up in the tool picker (the hammer icon); notes appear in the attach menu under resources.

Custom connectors are available on Free, Pro, Max, Team, and Enterprise plans; Free accounts are limited to one. On Team and Enterprise, an owner adds the connector under Settings → Organization Settings → Connectors first.

Connecting Claude Code

Claude Code supports remote MCP servers natively. To add it for your personal use across all projects:

claude mcp add --transport http --scope user nerdy https://nerdnoir.ai/api/mcp

To share it with everyone on a project (writes to .mcp.json in the repo root):

claude mcp add --transport http --scope project nerdy https://nerdnoir.ai/api/mcp

Or add it manually by creating .mcp.json at the project root:

{
  "mcpServers": {
    "nerdy": {
      "type": "http",
      "url": "https://nerdnoir.ai/api/mcp"
    }
  }
}

Once configured, run /mcp inside Claude Code to confirm the server shows as connected, and to complete the OAuth sign-in if you want the builder tools. Claude Code can then call collection tools mid-conversation — ask it to search the collection, pull in a concept, or list notes by tag.

Connecting Claude on the Web (claude.ai)

Custom remote MCP connectors are available on Free, Pro, Max, Team, and Enterprise plans. Free accounts are limited to one custom connector.

  1. Go to Settings → Connectors on claude.ai (or navigate directly to claude.ai/settings/connectors).
  2. Click + then Add custom connector.
  3. Enter a name (nerdy) and the server URL: https://nerdnoir.ai/api/mcp
  4. Click Add. The collection tools become available in your conversations.

Team and Enterprise accounts manage connectors under Settings → Organization Settings → Connectors.

Connecting ChatGPT

ChatGPT supports MCP natively via Developer Mode (beta, available on Plus, Pro, Business, Enterprise, and Education plans — not available on Free).

  1. Go to Settings → Apps and toggle on Developer mode in Advanced settings.
  2. Then go to Settings → Connectors → Create.
  3. Enter a name, description, and the endpoint URL: https://nerdnoir.ai/api/mcp

ChatGPT’s MCP client speaks streamable HTTP directly — no bridge or proxy needed. The integration is still rolling out and capabilities vary by plan; check OpenAI’s documentation for the current state.

Connecting Cursor

Cursor speaks streamable HTTP directly. Add the server to your MCP config using whichever scope fits:

Global (~/.cursor/mcp.json) — applies across all your projects:

{
  "mcpServers": {
    "nerdy": {
      "url": "https://nerdnoir.ai/api/mcp"
    }
  }
}

Project-level (.cursor/mcp.json in the repo root) — applies only to that project, overrides global:

{
  "mcpServers": {
    "nerdy": {
      "url": "https://nerdnoir.ai/api/mcp"
    }
  }
}

After saving, open Settings → MCP in Cursor and confirm nerdy is listed and enabled — Cursor gates MCP behind a per-server toggle, so a newly added entry doesn’t always turn itself on. If the tools don’t show up in a chat, run MCP: View Server Status from the command palette to check the connection.

Debugging with MCP Inspector

npx @modelcontextprotocol/inspector https://nerdnoir.ai/api/mcp

This launches a web UI where you can call each tool, browse resources, and inspect JSON-RPC messages.

Authentication

The read tools and the collection:// resources work with no sign-in. The builder, history, and Team tools read and write documents that belong to you, so they need an account. The server speaks OAuth 2.0 for those, and the flow runs like this:

  1. Your client discovers the OAuth server via /.well-known/oauth-protected-resource.
  2. It registers itself dynamically and opens a browser for you to log in.
  3. If you’re already logged into the site, the authorization step completes immediately.
  4. Your client receives a token and includes it in subsequent MCP requests.

A call to a builder tool with no token gets HTTP 401 with a WWW-Authenticate header, and that challenge is what triggers your client’s OAuth flow. A client that cannot do OAuth can still read the whole collection. Workshop slide decks are not part of the MCP server, signed in or not.

Rate Limits

Every caller is capped at 120 MCP requests per hour per IP. The cap runs ahead of the auth check, so signing in does not raise it. If you hit the ceiling the server returns HTTP 429 with a Retry-After header. Contact team@nerdnoir.com if your use case needs a higher ceiling.