infragate/capa

What it solves

CAPA simplifies the management of AI coding agents by consolidating scattered configurations—such as rules, MCP servers, and skills—into a single, version-controlled capabilities.yaml file. It eliminates the need to manually sync settings across different agents (like Cursor, Claude Code, and GitHub Copilot) and ensures that teammates cloning a repository get an identical agent setup via a lockfile.

How it works

CAPA acts as both a configuration manager and a runtime gateway. It reads a central YAML file and "fans out" the settings into the native formats required by 35+ different AI agents. At runtime, it provides a local MCP (Model Context Protocol) gateway that proxies requests from agents to upstream servers, allowing for lazy-loading of tools to reduce token usage and providing filtered endpoints for sub-agent isolation.

Who it’s for

Developers and teams who use multiple AI coding agents and want a standardized, shareable, and version-controlled way to manage the tools, rules, and capabilities these agents possess.

Highlights

• Unified Configuration: Write once in YAML and deploy to over 35 agents including Cursor, Claude Code, and Codex.
• MCP Gateway: A single local endpoint that proxies stdio, HTTP, and SSE servers with on-demand tool loading to reduce inference costs.
• Shadow Workspaces: The capa wrap command allows running agents in a separate workspace to prevent provider-specific configuration files from polluting the main repository.
• Observability: Includes a local Web UI for editing capabilities and a live activity feed to trace MCP calls and tool executions.
• Registry Support: Ability to browse and add skills and plugins from catalogs like the Cursor Marketplace and Claude plugins.