@ubgrade/cortex-mcp

Dynamic MCP bridge for Cortex Brain — gives Claude Code, Cursor, and any MCP-compatible client persistent memory across sessions.

No hardcoded tools. The platform controls what tools are available based on your API key tier and permissions.

Install

Claude Code

claude mcp add cortex -- npx @ubgrade/cortex-mcp

Set your API key:

export CORTEX_API_KEY=cortex_key_...

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "cortex": {
      "command": "npx",
      "args": ["@ubgrade/cortex-mcp"],
      "env": {
        "CORTEX_API_KEY": "cortex_key_..."
      }
    }
  }
}

Any MCP client

CORTEX_API_KEY=cortex_key_... npx @ubgrade/cortex-mcp

Configuration

| Environment Variable | Required | Default | Description | |---------------------|----------|---------|-------------| | CORTEX_API_KEY | Yes | — | Your API key from crewbase.cloud | | CORTEX_API_URL | No | https://api.ai-cortex.io | Platform API URL (use https://api.ai-cortex.eu for EU) |

How it works

1. Authenticates with your API key
2. Fetches available tools from the platform (GET /api/tools)
3. Registers each tool with the MCP SDK
4. Proxies tool calls back to the platform
5. Polls for tool updates every 60 seconds

Tools are loaded dynamically — when the platform adds new tools or changes your tier, they appear automatically.

Offline mode

After the first successful connection, tools are cached locally at ~/.cortex-mcp/tools-cache.json. If the platform is temporarily unreachable, the MCP server starts with cached tools.

ChatGPT Custom GPT

Generate an OpenAPI spec for ChatGPT Custom GPT actions:

CORTEX_API_KEY=cortex_key_... npx @ubgrade/cortex-mcp --openapi > openapi.json

Paste the generated spec into your Custom GPT's actions configuration.

CLI

npx @ubgrade/cortex-mcp              # Start MCP server (stdio)
npx @ubgrade/cortex-mcp --openapi    # Generate OpenAPI spec to stdout
npx @ubgrade/cortex-mcp --version    # Print version

License

MIT