# CAPA MCP server

> Writes one capabilities.yaml into 35+ AI coding agents' native config formats and proxies upstream MCP servers through a local gateway.

- Repository: https://github.com/infragate/capa
- Page: https://hostedmcp.org/servers/infragate/capa
- Official: yes
- Transport: —
- Stars: 747
- License: MIT

## FAQ

### What does CAPA do for AI coding agents?

It reads a single capabilities.yaml and writes skills, tools, rules, sub-agents, MCP servers, and plugins into each agent's native layout. It also runs a local MCP gateway in front of upstream servers.

### Which AI coding agents does CAPA support?

CAPA fans out to Cursor, Claude Code, Codex, Windsurf, GitHub Copilot, and over 35 other agents. `capa wrap` can also launch Claude Code, Codex, Cursor, Gemini CLI, OpenCode, and several others in a shadow workspace.

### How do I install CAPA?

On macOS and Linux use the curl install script; on Windows use the PowerShell one-liner. Then run `capa init` in your project to create capabilities.yaml.

### Does CAPA reduce token usage?

Yes. The MCP gateway loads tools on demand instead of front-loading the whole catalog, which the README reports as 19–40% fewer tokens across 150 trials.

### How do I run a configured tool from the terminal?

Every tool you define becomes a CLI command under `capa sh`. List tools with `capa sh`, then run a subcommand such as `capa sh brave search --query "…"`.
