MCP Server

Connect Rigour directly to your AI agents (Cursor, Claude Code, etc.) to enforce quality standards in real-time.

---

⚡ Quick Start

npx -y @rigour-labs/mcp

---

🔌 Integration Recipes

🤖 Claude Code (CLI)

claude mcp add rigour -- npx -y @rigour-labs/mcp

🖱️ Cursor

1. Go to Settings > Features > MCP.
2. + Add New MCP Server:
   • Name: Rigour
   • Type: command
   • Command: npx -y @rigour-labs/mcp

🛠️ Cline / Roo Code

Add this to your cline_mcp_settings.json:

{
  "mcpServers": {
    "rigour": {
      "command": "npx",
      "args": ["-y", "@rigour-labs/mcp"]
    }
  }
}

---

🛠️ Core Tools

Once connected, your AI agent will automatically use these tools:

Tool	Purpose
rigour_check	Runs all quality gates on your code (requires rigour.yml).
rigour_scan	Zero-config scan — auto-detects stack, no rigour.yml needed.
rigour_explain	Get detailed explanation of validation results.
rigour_learn	Teach Rigour patterns from failures — brain learns and reinforces over time.
rigour_recall	Retrieve stored engineering context and learned patterns.
rigour_status	Show current project status — score, patterns, config.

New Tools (v2.14+ – Frontier Model Support)

These tools enable supervision of Opus 4.6 agent teams and GPT-5.3 coworking mode:

Tool	Purpose
rigour_checkpoint	Report checkpoint status during long-running tasks.
rigour_agent_register	Register an agent in a multi-agent team session.
rigour_security_patterns	Pre-commit security pattern check.

// Agent team example
await mcp.call("rigour_agent_register", {
  cwd: "/project",
  agent_id: "agent-frontend",
  task_scope: ["src/ui/**", "src/components/**"]
});

// Checkpoint during long task
await mcp.call("rigour_checkpoint", {
  cwd: "/project",
  progress_pct: 50,
  files_changed: ["src/api/users.ts"],
  summary: "Implemented user endpoints"
});

rigour_run_supervised

Implements the full Supervisor Mode via MCP. Iteratively runs your agent command, checks gates, and provides fix packets until PASS or max retries.

// Example MCP call
rigour_run_supervised({
  cwd: "/path/to/project",
  command: "claude \"fix the auth bug\"",
  maxRetries: 3,  // default: 3
  dryRun: false   // skip command execution for testing
})

---

🏗️ How it Works: The Local Bridge

Even if your AI agent (like Cursor or Claude Code) is communicating with a model in the cloud, the Governance Interception happens entirely on your local machine.

The Handshake Architecture

graph TD
    subgraph "Your Machine (Local)"
        A["AI Agent (Cursor/Claude)"] -- runs --> B["Rigour MCP Server<br/>(local process)"]
        B -- writes events --> C[".rigour/events.jsonl<br/>(Shared File System)"]
        D["Rigour Studio<br/>(Local Dashboard)"] -- watches --> C
        D -- writes decision --> C
        B -- reads decision --> C
    end
    
    subgraph "Cloud"
        E["AI Model (Claude/GPT-4)"] -- instructs --> A
    end

The Synchronization Flow:

1. Local Spawn: When you add the Rigour MCP to your IDE, the IDE launches the server as a local process on your machine.
2. Shared State: Both the MCP server and the Studio UI point to the same .rigour folder in your project root.
3. Air-Gapped Arbitration:
   • The MCP server writes an "interception requested" log.
   • The Studio UI (via EventStream) detects this change instantly and pauses the agent.
   • Your local decision (Approve/Reject) is written back to the log.
   • The local MCP server sees your decision and returns the result to the AI agent.

🛡️ Zero-Telemetry & Local Trust

Rigour is intentionally local-first. Unlike other governance tools that require cloud connectivity, Rigour's interception layer is built on a "Local Bridge" architecture:

• Local Storage: All interception logs and human decisions are stored only in your .rigour/ folder.
• Air-Gapped Sync: The synchronization between your AI agent and the Governance Studio happens entirely on your machine via the local filesystem.
• Zero Push: We never push your source code, your command history, or your arbitration decisions to any external server.

This ensures that your project's engineering standards and command executions remain 100% private and under your total control.

🌐 Web vs. Desktop: The Governance Moat

Feature	Desktop (Cursor / Claude Desktop)	Web (Claude.ai / ChatGPT)
File Access	✅ Native (direct disk read/write)	❌ Sandboxed (no local disk access)
Trust Model	🛡️ Local Honest (100% Private)	☁️ Cloud-Mediated (requires tunnel)
Setup	npx @rigour-labs/mcp	Hosted Connector + Tunnel

Why doesn't the Web App see my local files?

For security, your browser cannot run shell commands or read your hard drive. To bridge Rigour to a web app, you would need to use a Secure Tunnel (like ngrok) to expose a local SSE server.

[!IMPORTANT] To maintain the Local Honest moat, we recommend using Rigour with Desktop AI Agents. If you must use the web version, look into our Remote MCP Server for the E2E-encrypted connectivity.

---

💡 Pro Tip

Always keep the Rigour Studio open on a second monitor while working with MCP agents. It provides a visual audit trail of everything the agent is doing.