https://github.com/infragate/capa
CAPA is a powerful package manager for AI agents that allows you to define skills and tools, manage credentials, and seamlessly integrate with MCP clients like Cursor and Claude
https://github.com/infragate/capa
agentic-workflow capabilities mcp mcp-server package-manager skills
Last synced: about 1 month ago
JSON representation
CAPA is a powerful package manager for AI agents that allows you to define skills and tools, manage credentials, and seamlessly integrate with MCP clients like Cursor and Claude
- Host: GitHub
- URL: https://github.com/infragate/capa
- Owner: infragate
- Created: 2026-02-08T20:29:26.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2026-03-06T23:34:58.000Z (4 months ago)
- Last Synced: 2026-03-07T04:45:18.994Z (4 months ago)
- Topics: agentic-workflow, capabilities, mcp, mcp-server, package-manager, skills
- Language: TypeScript
- Homepage: https://capa.infragate.ai
- Size: 462 KB
- Stars: 7
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# CAPA: Agentic Capabilities Manager
[](https://github.com/infragate/capa/releases/latest)
[](https://github.com/infragate/capa/actions/workflows/test.yml)
[](https://github.com/infragate/capa/actions/workflows/release.yml)
[](LICENSE)
[](https://github.com/infragate/capa/releases/latest)
**CAPA** is a powerful capabilities manager for AI agents that allows you to define skills and tools, manage credentials, and seamlessly integrate with agents like Cursor and Claude.
## Why CAPA?
AI agents need two things to be useful: *knowledge* of when and how to act, and the *ability* to actually do it. Most setups treat these separately — tools are wired up in one place, instructions scattered somewhere else. CAPA brings them together as a single unit called a **capability**.
- **Skills** provide the knowledge — markdown documents that give an agent context, instructions, and decision-making guidance for a specific task.
- **Tools** provide the ability — executable functions the agent calls to interact with the world: APIs, shell commands, file operations, and more.
A tool without knowledge leaves the agent unsure when to use it. Knowledge without tools leaves the agent unable to act. CAPA pairs them declaratively in a single `capabilities.yaml` file that you can version-control, share across a team, and reproduce on any machine.
## Features
- 🔌 Single MCP server that proxies only the necessary tools
- ⚡ Dynamic on-demand tool loading
- 🖥️ Expose shell commands as MCP tools
- 💻 Run any configured tool from the terminal with `capa sh`
- 🔑 Credential management via interactive UI or `.env` file
- 🛡️ Security controls (blocked phrases, character sanitization)
- 📦 Compatible with [skills.sh](https://skills.sh)
- 🤖 Supports Cursor and Claude plugin installation
- 🔒 Installation of skills and plugins from private repositories (GitHub and GitLab)
- 🧠 Self-improving agents
- 🎯 Default argument values for MCP tools
- 🔧 CLI prerequisite verification before installation
## Installation
**macOS and Linux:**
```bash
curl -LsSf https://capa.infragate.ai/install.sh | sh
```
**Windows:**
```powershell
powershell -ExecutionPolicy ByPass -c "irm https://capa.infragate.ai/install.ps1 | iex"
```
## Quick Start
### 1. Initialize your project
```bash
cd your-project
capa init
```
This creates a `capabilities.yaml` file where you define your agent's tools and skills.
### 2. Define your capabilities
```yaml
providers:
- cursor
skills:
- id: web-researcher
type: inline
def:
content: |
---
name: web-researcher
description: Search the web for information
---
Use the brave.search tool to find current information online.
requires:
- "@brave.search"
servers:
- id: brave
type: mcp
description: Brave web search
def:
cmd: npx -y @modelcontextprotocol/server-brave-search
env:
BRAVE_API_KEY: ${BraveApiKey}
tools:
- id: search
type: mcp
description: Search the web using Brave Search
def:
server: "@brave"
tool: brave_web_search
```
### 3. Install and launch
```bash
capa install
```
CAPA installs your skills, starts the capability server, and automatically registers with your MCP client (Cursor, Claude Desktop).
### 4. Run tools from the terminal
```bash
capa sh # list all available commands
capa sh brave # list brave subcommands
capa sh brave search --query "…" # run a tool directly
```
`capa sh` turns every configured tool into a CLI command. MCP tools are exposed as `server_name.tool_name` and grouped under their server ID in the CLI. Command tools appear at the top level (or under a custom `group`). Any unrecognised command is passed through to the OS shell.
## Documentation
For complete guides, examples, and API reference, visit:
**[https://capa.infragate.ai](https://capa.infragate.ai)**
## License
MIT