AI code review in Cursor.

How it works

AI Council ships with an MCP server (Model Context Protocol, the standard Cursor uses to connect external tools to its agent). Once registered, Cursor's agent can convene the Council on demand: it passes your current git diff to the review tools, and the Council returns per-agent votes with confidence scores, a debate transcript when agents disagree, and one final ruling: APPROVE, REVISE, or REJECT.

Your code goes directly from your machine to the model providers you configure with your own API keys. Nothing passes through AI Council's servers.

Prerequisites

• Node.js 18 or newer
• An AI Council license key
• An OpenAI or Google Gemini API key (at least one required; both recommended)

Step 1: Install the CLI

npm install -g @mugzie/ai-council

Step 2: Register the MCP server

Create an .mcp.json file in your project root (or your home directory to enable it everywhere):

{
  "mcpServers": {
    "ai-council": {
      "command": "ai-council",
      "args": ["mcp"],
      "env": {
        "AI_COUNCIL_LICENSE_KEY": "your-license-key",
        "OPENAI_API_KEY": "your-openai-key",
        "GEMINI_API_KEY": "your-gemini-key",
        "AI_COUNCIL_MODEL": "gpt-5.4-mini",
        "AI_COUNCIL_GEMINI_MODEL": "gemini-3.6-flash"
      }
    }
  }
}

Restart Cursor, then confirm the server is enabled under Cursor Settings → MCP.

Step 3: Ask for a review

In Cursor's agent chat, ask for a review in plain language:

"Have the AI Council review my current changes before I commit."

Cursor invokes the Council's review tool on your working-tree diff. The tools accept useful parameters you can request naturally:

• gitScope: review staged, unstaged, or all changes (defaults to auto)
• suggestions: agents include concrete code fixes with REVISE or REJECT votes
• deepAnalysis: file-by-file review with structured findings and project-rule awareness
• preContext: enrich reviews with AST-extracted, embedding-ranked code context

For example: "Run a deep analysis review of my staged changes with fix suggestions."

Or use the built-in terminal

The CLI works in Cursor's integrated terminal too, no MCP required:

# Review your working diff with formatted output
ai-council review --diff --pretty

# Review only staged changes, with fix suggestions
ai-council review --staged --suggestions

Why add a council to an editor that already has AI?

The model writing your code is a poor judge of its own output; it approves what it just reasoned itself into. AI Council supplies independent judgment: separate agents for security, performance, architecture, and pragmatism, each voting with a confidence score, with structured debate when they disagree. It supplements whatever model you code with; it doesn't replace it. See a real review transcript for what that looks like in practice.