Skip to main content

Quick Start

Get a cross-agent code review working in under 5 minutes.

1. Start a Server​

The fastest way to test is running a server directly in stdio mode:

# Run the Claude bridge directly
node dist/claude/server.js

# Or use the standalone binary
./exe/claude-mcp-server

For background HTTP mode (recommended for regular use):

# Start Claude bridge on port 8940
npx mcp-proxy --port 8940 -- ./exe/claude-mcp-server

2. Configure Your MCP Client​

Add the server to your MCP client configuration. For Claude Code, create or edit .mcp.json in your project root:

HTTP Mode (background service)​

{
  "mcpServers": {
    "claude_reviewer": {
      "type": "streamable-http",
      "url": "http://localhost:8940/mcp"
    }
  }
}

Stdio Mode (direct process)​

{
  "mcpServers": {
    "claude_reviewer": {
      "type": "stdio",
      "command": "claude-mcp-server"
    }
  }
}

3. Use It​

Once configured, your MCP client can call the bridge tools. For example, in Claude Code:

Use the claude_reviewer MCP to review this diff:

<paste your diff here>

The reviewer will return a structured JSON response:

{
  "verdict": "NEEDS_REVISION",
  "issues": [
    {
      "severity": "major",
      "description": "SQL query is vulnerable to injection",
      "recommendation": "Use parameterized queries instead of string interpolation"
    }
  ],
  "suggestions": [
    "Consider adding input validation at the API boundary"
  ]
}

4. Add More Agents​

Add Codex and Copilot bridges for cross-agent validation:

{
  "mcpServers": {
    "claude_reviewer": {
      "type": "streamable-http",
      "url": "http://localhost:8940/mcp"
    },
    "codex": {
      "type": "streamable-http",
      "url": "http://localhost:8941/mcp"
    },
    "copilot": {
      "type": "streamable-http",
      "url": "http://localhost:8945/mcp"
    }
  }
}

Now you can ask one agent to get a second opinion from another:

Send this diff to the codex MCP for code review, then compare
with the claude_reviewer's assessment.

Next Steps​