OnCall Runbook

You run an OnCall Runbook MCP Server that centers around the rb.answer tool to deliver fast, structured, and reliable runbook answers. It supports intelligent intent understanding, structured outputs with risk alerts, and graceful degradation when offline, making it practical for on-call troubleshooting and incident response.

How to use

Use an MCP client to invoke the rb.answer tool and get enhanced, structured responses for operational questions. You will see a multi-part reply that includes a concise summary, relevant citations, risk assessments, safe recommended operations, and metadata about the answer quality and LLM availability. You can use online mode when an LLM is available for smarter summaries, or offline mode when you want structured references without LLM input. In degraded mode, the system falls back to safe, structured references with guidance for escalation.

How to install

Prerequisites: ensure you have Node.js and npm installed on your system. You will set up the MCP server locally and verify it with tests before running in production.

# 1. Install dependencies for the MCP server project
npm install

# 2. Run unit/integration tests to verify setup
npm test

# 3. Run specific integration tests to verify rb.answer behavior
node tests/integration/us1-tools.spec.mjs

Additional configuration and running guidance

The MCP server exposes a local runtime via a standard node process. The runtime command to start the server is shown in the embedded client configuration example. Use the following runtime command to start the server locally.

{
  "mcpServers": {
    "oncall_runbook": {
      "type": "stdio",
      "name": "oncall_runbook",
      "command": "node",
      "args": ["server.mjs"]
    }
  }
}

Claude Desktop and client integration

If you use Claude Desktop or a compatible MCP client, configure the client to connect to the local server using the stdio configuration shown above. The client will call rb.answer with your questions and receive a structured, enhanced response.

Experiment and run modes

• Online Mode: LLM is available; you receive smart summaries and analysis.
• Offline Mode: No LLM; you receive structured citations and procedures.
• Degraded Mode: LLM is temporarily unavailable; the server gracefully degrades with safe, actionable guidance.
• No Results: If nothing matches, you receive escalation suggestions to help you proceed.

Notes on performance and testing

The server includes query caching, concurrency management, and response-time monitoring to ensure responsive behavior under load. Tests cover core functionality, citation formatting, and end-to-end answer generation in both LLM-on and LLM-off scenarios.

Available tools

rb.answer

A single, central tool that handles intelligent Q&A, intent understanding, structured output, risk assessment, and degraded-mode behavior for runbooks.