cognitive-core-mcp

Standalone MCP server (stdio) that exposes Cognitive Core Rooms as tools to any MCP-compatible client — Claude Code, Cherry Studio, Continue, Zed, Cursor.

A thin HTTP bridge: ~300 lines of Python, no state, no memory of its own.

Install

pip install --user cognitive-core-mcp

Or no-pip flavour:

mkdir -p ~/.local/bin
curl -fsSL https://raw.githubusercontent.com/mocartlex-wq/cognitive-core/main/mcp-wrapper/cognitive_core_mcp.py \
  -o ~/.local/bin/cognitive_core_mcp
chmod +x ~/.local/bin/cognitive_core_mcp
pip install --user mcp httpx

Configure Claude Code

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "cogcore-rooms": {
      "command": "cognitive_core_mcp",
      "env": {
        "COGCORE_URL":      "https://your-server.example",
        "COGCORE_AGENT_ID": "alice",
        "COGCORE_ROOM_KEY": "your-default-room-api-key"
      }
    }
  }
}

Restart Claude Code. You'll see cognitive_room_* tools in the picker.

Tools

Env vars

Development

git clone https://github.com/mocartlex-wq/cognitive-core
cd launch/mcp-wrapper
python -m venv .venv && . .venv/bin/activate
pip install -e .
cognitive_core_mcp   # stdio MCP — talk via JSON-RPC framing

For interactive testing without an MCP client:

COGCORE_LOG=debug python -c "
import asyncio
from cognitive_core_mcp import call_tool
print(asyncio.run(call_tool('cognitive_room_health', {})))
"

License

MIT.