MCP Playbook

What it does

MCP Playbook connects AI coding assistants to a structured documentation workflow. When you start a session, the AI calls init_playbook to receive instructions, then uses the server's tools to create ADRs, specs, changelogs, and runbooks inside your project directory — plus upload conversation logs and prompts to GitHub repositories maintained by Dwarves Foundation.

Tools

• init_playbook — Returns instructions so the AI knows how to use the server at session start
• create_spec — Creates a numbered spec file (PRD, RFC, architectural plan) in docs/specs/
• create_adr — Creates a numbered Architectural Decision Record in docs/adr/
• create_changelog — Creates a detailed changelog entry in docs/changelog/
• distill_project_runbook — Synthesises existing docs into an updated docs/runbook.md
• save_and_upload_chat_log — Saves conversation history locally and uploads it to dwarvesf/prompt-log on GitHub
• search_runbook — Fuzzy-searches the dwarvesf/runbook GitHub repo and returns top 5 matches with full content
• search_prompts — Fuzzy-searches the dwarvesf/prompt-db GitHub repo
• suggest_runbook — Opens a PR in dwarvesf/runbook with a new runbook entry
• sync_prompt — Syncs an LLM prompt to the dwarvesf/prompt-db GitHub repo
• think — Logs internal reasoning without making file or API changes

Installation

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "mcp-playbook": {
      "command": "npx",
      "args": ["-y", "@dwarvesf/mcp-playbook@latest"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "your_github_token_here"
      }
    }
  }
}

Also add this to your Cursor rules, Cline rules, or Windsurf rules so the AI initialises correctly:

ALWAYS run the `init_playbook` tool before starting to get instructions on how things should be done and documented.

Supported hosts

Confirmed working with Claude Desktop, Cursor, Cline, and Windsurf via stdio transport.