Cerina MCP Server

MCP (Model Context Protocol) server that exposes the CBT Clinical Review Multi-Agent System as tools for AI assistants like Claude Desktop.

Features

• create_cbt_exercise: Create CBT exercises using the multi-agent workflow
• list_exercises: List approved exercises from the database
• get_session_status: Check status of exercise creation sessions
• approve_exercise: Approve exercises awaiting human review

Prerequisites

1. Python 3.13+
2. uv package manager
3. Cerina Foundry backend running at http://localhost:8000

Installation

cd mcp
uv sync

Claude Desktop Configuration

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

Daily Use

{
  "mcpServers": {
    "cerina-mcp": {
      "command": "uvx",
      "args": ["cerina-foundry"]
    }
  }
}

Development

{
  "mcpServers": {
    "cerina-mcp": {
      "command": "uv",
      "args": [
        "--directory",
        "/Users/jagjeevankashid/Developer/cerina/mcp",
        "run",
        "cerina-foundry"
      ],
      "env": {
        "CERINA_BACKEND_URL": "http://localhost:8000/api/v1",
      }
    }
  }
}

Environment Variables

Usage Examples

Once configured in Claude Desktop:

Create an Exercise

"Ask Cerina Foundry to create a sleep hygiene protocol for patients with insomnia"

List Exercises

"Show me the approved CBT exercises for anxiety"

Check Status

"What's the status of session abc123?"

Approve Pending Exercise

"Approve the exercise in session xyz789"

Development

Run the server manually:

uv run cerina-foundry

Test with MCP inspector:

npx @modelcontextprotocol/inspector uv --directory . run cerina-foundry