Runbook MCP Server

🚀 Runbook.AI Runbook Server

Runbook.AI Runbook Server enables you to run your own runbooks via Claude Desktop.

• Execute operational runbooks (e.g., deploy services, upgrade Kubernetes clusters).
• Execute manual test plans (e.g., create a new EC2 instance, SSH into it, and run integration tests there).

Please watch the demo video below to see how it works!

🚀 Quick Start

Step 1: Start the MCP server using Claude Desktop.

Create config.yaml. Put the following configuration into claude_desktop_config.json.

{
  "mcpServers": {
    "runbook": {
      "command": "uv",
      "args": [
        "--directory",
        "<ABSOLUTE_PATH>/runbook-mcp-server",
        "run",
        "runbook_server.py"
      ]
    }
  }
}

Step 2: Create a runbook

You can use Markdown files or text files as runbooks. See Examples. We support environment/variable substitution. Define env.yaml.

env:
  key: value

• {env[key]} is replaced with value.
• {var.name} is replaced with the variable value. Rebuild the index when updating the file. You can also use the create_runbook tool to create a new runbook. Here is an example prompt:

Example 1

Create a new runbook:

- name: list_pods
- content: List all pods in all namespaces, find the unready pods, and send this information to Slack.

You don't need to explain the content. Just pass it to the tool.

Step 3: Execute the runbook

To run a runbook, follow these steps:

1. Click "Attach from MCP" in Claude Desktop.
2. Select get_runbook_as_prompt (from the integration list).
3. Pass the name of the runbook you wish to execute. Also, specify variables using a comma-separated list (e.g., var1=key1,var2=key2).
4. Submit the generated prompt.

Then Claude Desktop will communicate with other MCP servers to execute the runbook.

💻 Usage Examples

Basic Usage

The above steps show the basic usage of creating and executing a runbook.

Advanced Usage

You can further customize your runbooks by using environment/variable substitution and creating more complex runbook content.

📦 Example MCP Servers for Running Runbooks

• GitHub
• Slack
• DesktopCommanderMCP
• Web search and browser automation (Link)
• Kubernetes. There are several implementations (e.g., mcp-k8s-go)

📚 Documentation

Development Plan

• Instead of saving runbook content in a database, save it as files. This facilitates easy editing. People can also use GitHub for version control.
• Runbook template X, instantiated with given inputs.
  • This feature may not be necessary. Users can directly pass additional prompts at runtime.
• Save execution logs (for auditing and optimization).
  • Remove confidential information.
  • Also pass past logs into the runbook prompt if it helps with better execution.
• Approval process.
  • Add the tool request_approval.
  • This sends a message to a Slack channel.
  • Then the Runbook.AI MCP server monitors the channel. If someone replies (yes/no), it continues or returns an error.
• Better runbook search.
  • Exact name matching is not ideal.
• Register with Smithery.

Possible Work Items with Uncertain Feasibility

Note: Claude Desktop does not support "sampling". This imposes some limitations.

• Sub-runbooks and reusable execution blocks.
• REST endpoints with a front-end (for managing runbooks).
• Ability to edit runbooks (with version control).
• Display the execution plan (drill) of runbook X.
  • Limit MCP servers and tools.
• Convert previous conversations into runbooks.
• Fine-tuning.
• Ability to optimize runbooks through successful execution. If there is a successful execution, save it as an example and pass it to Claude.
• Periodic execution.
• Ability to provision environments (VMs, Docker) to run MCP servers.