Show HN: Reusable MCP Playbooks for AI Agents (github.com)

Director announced “MCP Playbooks” — a portable, local-first system that packages tools, prompts and configuration into declarative YAML playbooks to give AI agents reusable skills. Playbooks can be swapped with one click (currently supporting Claude Code/Desktop, Cursor and VSCode), shared via version control, and enforced at startup so clients only see the mapped tools and prompts. The project positions itself as “Terraform for AI agents”: a gateway that sits between agents and MCP servers, providing unified OAuth, strong isolation, centralized JSON logging, tool filtering, and observability while remaining transparent to clients (no extra tokens required). Technically, playbooks are MCP-compliant and declare servers (HTTP or stdio), included tool lists, and prompts; examples include GitHub and Slack integrations. Director ships as a CLI, UI (studio), Typescript SDK, Docker image and sandbox, with a one-line installer (curl … | sh) and a programmatic Gateway.start API for advanced workflows. The system enforces client<>playbook mappings on startup, supports registry-driven server entries, and exposes full management commands (create/connect/auth/list/call tools, edit prompts). Distributed as an open monorepo (AGPL v3), Director aims to make agent capabilities reproducible, shareable, and secure for teams building multi-tool AI workflows.