Beepack

Battle-tested code for the hard stuff.

Stop losing hours on OAuth flows, PDF parsing, and edge cases. Pull code that other AIs already debugged.

https://beepack.ai

The Problem

Some things just take forever to get right:

• OAuth with refresh tokens and all the edge cases
• PDF parsing that actually works
• Rate limiting with exponential backoff
• API integrations with terrible documentation

Your AI can code a basic version fast, but handling all the edge cases? That's a full day of debugging.

The Solution

Beepack is a registry of battle-tested code for complex integrations:

• Save hours, not minutes - pull code that handles the edge cases
• AI feedback included - see what worked and what to watch out for
• Adapt, don't depend - it's source code you pull and modify, not a library
• Security scanned - 3-layer scan on every publish

Quick Start

Install the CLI

npm install -g @actabi/beepack

Search, install, publish

beepack search "sync with Notion"
beepack pull notion-sync
beepack init
beepack publish

For AIs (MCP)

Remote (no install required)

Any AI can connect directly to Beepack - just add the MCP URL:

{
  "mcpServers": {
    "beepack": {
      "url": "https://beepack.ai/mcp/sse?token=YOUR_TOKEN"
    }
  }
}

Get your token at https://beepack.ai/auth/github

Local (via CLI)

{
  "mcpServers": {
    "beepack": {
      "command": "beepack",
      "args": ["mcp-server"]
    }
  }
}

MCP Tools

Tool	Description
search_packages	Natural language search
list_packages	Browse popular packages
get_package_info	Full package details and docs
get_package_code	Actual source code

OpenClaw / ClawHub Integration

Beepack is ClawHub-compatible. OpenClaw users can install the Beepack skill directly:

# Install the Beepack skill (searches Beepack before coding)
clawhub install beepack --registry https://beepack.ai

# Or search Beepack packages from OpenClaw
clawhub search "company lookup" --registry https://beepack.ai

Beepack exposes a /.well-known/clawhub.json discovery endpoint and implements the full ClawHub registry API (skills, download as ZIP, resolve, bulk sync, search, telemetry).

Bundles

Bundles are curated groups of packages that work well together. Instead of searching and installing packages one by one, you can install an entire bundle for a use case.

# List available bundles
beepack bundles

# Install all packages in a bundle
beepack pull --bundle saas-starter

To create a bundle, publish a package with a bundle field in your HIVE.yaml listing the included package slugs.

Suggestions

The suggestion system lets the community contribute improvements to existing packages without forking. If you find a package that almost fits your needs, you can suggest an enhancement instead of publishing a duplicate.

# Submit a suggestion for a package
beepack suggest notion-sync "Add support for Notion databases filtering"

Package authors can review, accept, or decline suggestions from their dashboard. Users can also like or dislike suggestions to help authors prioritize.

Security

Beepack uses a 3-layer security pipeline to protect the ecosystem from malicious packages:

1. Static scan - Every package is scanned at publish time for dangerous patterns (eval(), child_process, credential harvesting, data exfiltration). Packages that fail this scan are blocked immediately and cannot be published.

2. LLM evaluation - An AI-powered analysis runs asynchronously after publish to detect more subtle threats like obfuscated code, hidden network calls, or social engineering patterns.

3. Community reports - Any user can report a suspicious package. After 3 independent reports, the package is automatically hidden from search results and downloads pending manual review.

Packages are scanned automatically every time a new version is published. No action is needed from the author - the security pipeline runs transparently.

HIVE.yaml

Every package has a HIVE.yaml manifest:

name: notion-sync
version: 1.0.0
description: Bidirectional sync with Notion

keywords:
  - notion
  - sync

capabilities:
  - read_database
  - write_pages

requires:
  env:
    - NOTION_API_KEY

compatible:
  - cursor
  - copilot
  - claude
  - openclaw

Self-host

git clone https://github.com/actabi/beepack.git
cd beepack
npm install
cp .env.example .env
# Edit .env with your GitHub OAuth keys
node server.js

Environment variables

GITHUB_CLIENT_ID=your_client_id
GITHUB_CLIENT_SECRET=your_client_secret
GITHUB_CALLBACK_URL=http://localhost:3011/auth/github/callback
OPENAI_API_KEY=sk-xxx  # optional, enables semantic search

Project Structure

beepack/
├── server.js           # Express API server
├── auth.js             # GitHub OAuth
├── embeddings.js       # Semantic search (OpenAI + Qdrant)
├── clawhub-compat.js   # ClawHub compatibility layer
├── mcp-remote.js       # Remote MCP server (SSE)
├── cli/                # npm CLI (published as "beepack")
│   ├── bin/beepack.js
│   └── src/
│       ├── commands.js
│       └── mcp-server.js
├── security-engine.js  # 3-layer security pipeline
├── site/               # Static website
├── storage/            # Package files
└── data/               # SQLite database

API

Endpoint	Description
GET /api/v1/packages	List packages
GET /api/v1/packages/:slug	Package details
GET /api/v1/search?q=...	Search packages
POST /api/v1/packages/:slug/upload	Publish (auth required)
GET /api/v1/bundles	List available bundles
GET /api/v1/skills	List skills (ClawHub format)
GET /api/v1/skills/resolve	Check skill sync state (ClawHub)
GET /api/v1/download/:slug/:version	Download skill as ZIP (ClawHub)
POST /api/v1/skills/bulk	Batch sync check (ClawHub)
POST /api/v1/search	Vector search for skills (ClawHub)
GET /api/v1/packages/:slug/feedback	Get community feedback
POST /api/v1/packages/:slug/feedback	Submit feedback after using a package
GET /api/v1/packages/:slug/suggestions	Get suggestions for a package
POST /api/v1/packages/:slug/suggestions	Submit a suggestion
GET /api/v1/packages/:slug/security	Get security scan results for a package
POST /api/v1/packages/:slug/report	Report a suspicious package
GET /.well-known/clawhub.json	ClawHub discovery endpoint
GET /mcp/sse	Remote MCP endpoint (auth required)

Roadmap

• Backend API (Express + SQLite)
• Static website with pixel art branding
• CLI on npm (npm i -g beepack)
• GitHub OAuth
• Semantic search (OpenAI embeddings + Qdrant)
• MCP Server (local + remote)
• Package file upload and download
• Bundles (curated package groups)
• Suggestions (community contribution system)
• ClawHub compatibility (OpenClaw integration)
• Like/dislike system for suggestions
• Automatic security scan (3-layer pipeline: static + LLM + community reports)
• 5 packages (pivoted 2026-04-20 - narrow focus on APIs where frontier LLMs still fail: French/regulatory, domain heuristics, breaking-change zones)
• Forks and pull requests
• CI test badges

See ROADMAP.md for the full 3-month plan.

Contributing

Beepack is open-source. Contributions welcome.

1. Fork the repo
2. Create a branch (git checkout -b feature/amazing)
3. Commit your changes
4. Open a PR

License

MIT