Connector

claude.ai and Claude Desktop

1. 1In Claude, open Customize, then Connectors.
2. 2Click + Add, then Add custom connector.
3. 3Name it cowpath, paste the URL and continue.
4. 4Sign in to cowpath and pick your workspace.
5. 5In a chat, open +, then Connectors, and switch cowpath on.

Claude Desktop uses the same connectors as claude.ai, so adding it once is enough. On a Team or Enterprise plan, an owner adds cowpath first under Organization settings → Connectors; then everyone connects it under Customize → Connectors.

ChatGPT

ChatGPT takes custom connectors on Business, Enterprise and Edu workspaces, on the web. An admin may need to allow them first.

1. 1Open chatgpt.com/plugins, click +, then Add custom MCP server.
2. 2Name it cowpath, paste the URL and choose OAuth.
3. 3Create it, then sign in to cowpath and pick your workspace.
4. 4In a chat, type @ and pick cowpath.

Claude Code and Codex

Use the CLI instead: cowpath init adds hooks, so runs are kept in the background without your agent calling anything, and cowpath login shares them with your team. See the install guide.

To use only the connector:

claude mcp add --transport http --scope user cowpath https://cowpath.sh/mcp

Then run /mcp in Claude Code to sign in.

codex mcp add cowpath --url https://cowpath.sh/mcp
codex mcp login cowpath

Editors and other agents

Add cowpath to the editor's MCP settings. Each one opens the browser to sign in the first time.

Cursor

{
  "mcpServers": {
    "cowpath": {
      "url": "https://cowpath.sh/mcp"
    }
  }
}

VS Code with Copilot

{
  "servers": {
    "cowpath": {
      "type": "http",
      "url": "https://cowpath.sh/mcp"
    }
  }
}

Gemini CLI

{
  "mcpServers": {
    "cowpath": {
      "httpUrl": "https://cowpath.sh/mcp"
    }
  }
}

Windsurf

{
  "mcpServers": {
    "cowpath": {
      "serverUrl": "https://cowpath.sh/mcp"
    }
  }
}

The file is ~/.codeium/windsurf/mcp_config.json, or ~/.config/devin/mcp_config.json since Windsurf became Devin Desktop.

Any other agent that speaks MCP works the same way: a remote server at the URL above, signed in with OAuth.

What your agent does

You keep working as before. When a task ends in something you'd make again:

1. 1Before it starts, your agent finds the matching process, or creates one.
2. 2It follows that process's playbook.
3. 3When the result is final, it saves the run: the final version and your corrections.
4. 4When you approve or reject the result, it records your thumb. It never rates on its own.

One-off questions are skipped, and the full chat is never stored. Every fifth run, your agent rewrites the playbook. The docs explain the details.

Remove it

Remove cowpath where you added it. Your workspace stays; delete it in the app if you want it gone.