MCP / AI Agents

dops includes an MCP server that exposes runbooks as tools for AI agents like Claude.

---

Quick Setup with Claude Code

Add to your Claude Code MCP config:

{
  "mcpServers": {
    "dops": {
      "command": "dops",
      "args": ["mcp", "serve"]
    }
  }
}

Claude can now discover and execute your runbooks as tools.

---

Docker

Run the MCP server in a container:

docker run -i --rm \
  -v ~/.dops:/data/dops \
  ghcr.io/rundops/dops:latest

The container uses DOPS_HOME=/data/dops and runs dops mcp serve --transport stdio by default.

For HTTP transport:

docker run -d --rm \
  -v ~/.dops:/data/dops \
  -p 8080:8080 \
  ghcr.io/rundops/dops:latest \
  dops mcp serve --transport http --port 8080

---

Transports

Transport	Flag	Use Case
stdio	--transport stdio	Claude Code, local AI agents
http	--transport http --port 8080	Remote agents, web integrations

HTTP transport includes gzip compression and supports SSE for streaming.

---

Risk Control

Limit which runbooks are exposed by risk level:

dops mcp serve --allow-risk medium

This hides high and critical runbooks from AI agents.

---

How It Works

Tools

Each runbook becomes an MCP tool. The tool name is the runbook ID (e.g., infra.health-check). Parameters are mapped to the tool's JSON Schema input.

• Sensitive parameters are excluded from the schema and noted in the tool description.
• High-risk runbooks require a _confirm_id parameter matching the runbook ID.
• Critical-risk runbooks require a _confirm_word parameter set to CONFIRM.

Resources

The MCP server exposes two schema resources:

URI	Description
dops://schema/runbook	Full runbook.yaml schema reference
dops://schema/shell-style	POSIX shell scripting guide

Per-runbook resources are also available at dops://runbook/<id>.

Prompts

The create-runbook prompt guides AI agents through creating a new runbook with the correct YAML schema and shell template.

---

List Available Tools

dops mcp tools
dops mcp tools --allow-risk medium