MCP Integration

What is MCP?

The Model Context Protocol (MCP) is an open standard that lets AI clients call tools on external servers. CHAOS exposes its full capability surface via MCP:

• 134 tools — every agent, skill, and resource operation
• 19 resources — project context, session history, agent state, knowledge graph
• 8 workflow prompts — pre-built prompts for common pipelines

Starting the MCP Server

stdio (for local editors)

cw mcp

stdio transport is the default. The server reads from stdin and writes to stdout — ideal for Claude Desktop, local VS Code Copilot, and Cursor.

Streamable HTTP (for persistent hosting)

cw mcp --transport streamable-http
# Default port: 8000
# Override: cw mcp --transport streamable-http --port 9000

HTTP transport is for Docker deployments and shared team instances.

Docker (recommended for teams)

docker compose up -d
# MCP server available at http://localhost:8000

Client Configuration

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "chaos": {
      "command": "cw",
      "args": ["mcp"]
    }
  }
}

VS Code Copilot

Add to .vscode/mcp.json in your project:

{
  "servers": {
    "chaos": {
      "type": "stdio",
      "command": "cw",
      "args": ["mcp"]
    }
  }
}

Cursor

Add to Cursor settings under MCP servers:

{
  "chaos": {
    "command": "cw",
    "args": ["mcp"]
  }
}

HTTP clients (Windsurf, remote)

{
  "chaos": {
    "type": "http",
    "url": "http://localhost:8000/mcp"
  }
}

Verifying the Connection

List all available tools:

cw mcp --list-tools      # 134 tools
cw mcp --list-resources  # 19 resources
cw mcp --list-prompts    # 8 workflow prompts

Once connected, all CHAOS agents appear as callable tools in your editor. You can invoke any agent, run any skill, or query any context resource directly from the chat interface.

Example Tool Calls (from your editor)

Once CHAOS is connected as an MCP server, you can ask your AI assistant:

• "Run the security agent on src/auth/"
• "Generate tests for the payments module"
• "Show me the recent agent session history"
• "Run the pre-commit pipeline"
• "What does the knowledge graph say about our database layer?"

The assistant calls the corresponding CHAOS MCP tool and returns the result inline in your conversation.

Next Steps

• Providers — Configure your LLM providers
• Pipelines — Build and run agent pipelines
• Agents — Browse the full agent catalog