Installation

Install via npm:

npm install -g cog-mcp

Configuration

Add to your opencode.json config file:

{
  "mcpServers": {
    "cog-mcp": {
      "command": "cog-mcp",
      "args": ["--path", "/path/to/your/codebase"]
    }
  }
}

Usage

Once configured, you can use semantic code search directly in your AI assistant conversations.

Cog

Semantic code search for AI assistants. Powered by Nomic embeddings on Apple Silicon.

What it does

• Search code by meaning - Find functions that "handle authentication" not just keyword matches
• Analyze structure - Extract functions, classes, and dependencies from any file
• Generate embeddings - 768-dim vectors on Metal GPU (M1/M2/M3)

Requirements

• macOS with Apple Silicon
• Python 3.12+
• Node.js 18+
• uv

Install

git clone https://github.com/vitorcalvi/cog.git
cd cog

# Python engine
cd packages/core && uv sync

# MCP server
cd ../mcp && npm install && npm run build

Index your code

cd packages/core
uv run cog-index ./your/project

Connect to OpenCode

Edit ~/.config/opencode/opencode.json:

{
  "mcp": {
    "cog": {
      "type": "local",
      "command": ["node", "/ABSOLUTE/PATH/TO/cog/packages/mcp/dist/index.js"],
      "environment": {
        "COG_CORE_DIR": "/ABSOLUTE/PATH/TO/cog/packages/core",
        "COG_DB_PATH": "/ABSOLUTE/PATH/TO/cog/packages/core/cog_memory"
      },
      "enabled": true
    }
  }
}

Replace /ABSOLUTE/PATH/TO with your actual path.

Use in OpenCode

> Search my codebase for user authentication
> Analyze the structure of src/api.py
> Find database connection functions

Tools

Test it works

Run tests:

cd packages/core
uv run pytest
cd packages/mcp
npm test

License

MIT © Vitor Calvi