capa

Why we included this project

Keeping agent configuration in sync across a team is usually a mess: CLAUDE.md, .cursor/rules, AGENTS.md, MCP JSON, hooks, and skill folders all drift apart, and nothing is pinned. capa collapses that into one capabilities.yaml file checked into the repo. Run capa install and it writes the declared skills, tools, rules, sub-agents, MCP servers, hooks, and plugins into Cursor, Claude Code, Codex, Windsurf, GitHub Copilot, and dozens of other agents in each tool's native format, with SHAs pinned so a fresh clone reproduces the exact setup. At runtime it doubles as a local MCP gateway: every agent talks to one endpoint, while capa proxies upstream servers, loads tools only when called, and restricts what each sub-agent can reach. That matters for teams running several agents, and for anyone who wants to trim MCP context and token usage without giving up model quality.