MCP Second Opinion Server

You have access to an MCP Second Opinion Server that uses Gemini technology to provide AI-assisted code reviews and debugging guidance. It supports multi-turn sessions, autonomous tool use, Playwright-based UI debugging, and costs tracking, making it a practical assistant for tackling challenging coding tasks and debugging workflows.

How to use

Start by running the MCP server locally or connect to its HTTP endpoint from your MCP client. Use a multi-turn session to maintain context across related debugging steps, capture screenshots with Playwright when needed, and send code snippets, issue descriptions, and images to Gemini-powered analysis for iterative improvements.

How to install

Prerequisites: Python 3.11+, a Conda environment (recommended) or pip, and Git.

1. Clone the project and navigate into the MCP directory.

2. Set up the MCP Second Opinion server environment.

3. Install required dependencies and start the server as shown.

Configuration and start commands

Use the following commands and files to configure and run the MCP Second Opinion server. The server can be started locally with Python and is designed to be consumed by Claude Code via an MCP client.

# 1) Prepare the environment
cd mcp-second-opinion
cp .env.example .env
# Add your GEMINI_API_KEY to .env

# 2) Start the server locally
python src/server.py

MCP integration for Claude Code

Add an MCP connection in your Claude Code config to route requests to the Second Opinion server either via HTTP or as a local stdio process.

{
  "mcpServers": {
    "second-opinion": {
      "type": "http",
      "url": "http://127.0.0.1:8080/sse",
      "args": []
    }
  }
}

Environment variables

Provide the Gemini API key and any other required values to enable authentication and access to the Gemini services.

{
  "GEMINI_API_KEY": "YOUR_API_KEY"
}

Available tools

get_code_second_opinion

Comprehensive code analysis tool that reviews code snippets and provides actionable recommendations.

health_check

Check server status and health metrics to ensure the MCP endpoint is responsive.

create_session

Start a multi-turn consultation session to maintain context across messages.

consult

Send messages within an active session to request code reviews and guidance.

get_session_history

Retrieve the history of a given session for reference and audit.

close_session

End a session and provide a summary of the interaction and cost implications.

list_sessions

List all active and past sessions for quick navigation.

approve_fetch_domain

Approve a domain for safely fetching external documentation or resources.

revoke_fetch_domain

Revoke approval for a previously trusted domain.

list_fetch_domains

List all domains that have been approved for fetching