fdmtl/director

What it solves

Director provides a centralized way to manage and deploy "playbooks"—sets of tools, prompts, and configurations—to AI agents. It eliminates the need to manually configure individual Model Context Protocol (MCP) servers for every agent, allowing users to switch between different skill sets based on the task or environment.

How it works

Director acts as a transparent gateway service that sits between AI agents (like Claude, Cursor, or VSCode) and MCP servers. It bundles MCP tools and prompts into portable "playbooks" that can be shared via a single endpoint. It handles the underlying connectivity, authentication, and tool filtering to ensure agents only receive the capabilities they need for a specific context.

Who it’s for

Developers and AI power users who use MCP-compliant agents and want to organize their tools and prompts into reusable, task-specific skill sets that can be rapidly switched.

Highlights

• Playbooks: Group tools, prompts, and config into portable skills for different tasks.
• 1-Click Integration: Rapidly connect playbooks to Claude Code, Claude Desktop, Cursor, and VSCode.
• Unified OAuth: Centralized authentication for OAuth-based MCP servers across all agents.
• Tool Filtering: Preserve agent context by selecting only the necessary tools for a task.
• Local-First: Designed to run on your own machine or infrastructure for better control.
• Observability: Centralized JSON logging to track agent actions.