# Unbrowse

> Unbrowse is an open-source CLI tool that reverse-engineers the internal APIs behind any website, allowing AI agents to make direct API calls instead of automating headless browsers. It reduces page interaction time from 5-30 seconds to 50-200 milliseconds and cuts token usage from ~8,000 to ~200 tokens per action. Skills discovered by one agent are shared in a public registry for all agents to reuse.

## Install

One-command setup that installs browser assets, configures your agent host, and starts the local server:

```
npx unbrowse setup
```

Install globally for repeat use:

```
npm install -g unbrowse
unbrowse setup
```

If your agent host uses skills (OpenClaw, Claude Code, Cursor):

```
npx skills add unbrowse-ai/unbrowse
```

## How It Works

1. Your agent calls `unbrowse resolve --intent "search stays" --url "https://airbnb.com"`.
2. Unbrowse checks the shared skill registry for a matching API skill.
3. If no skill exists, it opens a local browser, captures network traffic, and reverse-engineers the site's internal APIs.
4. The discovered API endpoints are published to the registry so all agents benefit.
5. Your agent gets clean JSON data via direct API call — no DOM, no rendering, no HTML parsing.

## Key Pages

- [Homepage](https://www.unbrowse.ai): Product overview, demo, and installation instructions
- [Skill Registry](https://www.unbrowse.ai/search): Search discovered API skills by intent or domain — the shared marketplace of reverse-engineered APIs
- [Agent Dashboard](https://www.unbrowse.ai/dashboard): View your discovered skills, execution stats, and API key management
- [Skill Documentation](https://www.unbrowse.ai/skill.md): Complete technical documentation for the unbrowse skill — all CLI commands, API endpoints, auth flows, and rules
- [Privacy Policy](https://www.unbrowse.ai/privacy): Details on what data is shared to the registry vs kept local on your machine
- [Terms of Service](https://www.unbrowse.ai/terms): Usage terms for unbrowse and the skill registry

## API

Base URL: `https://beta-api.unbrowse.ai`

### Core Endpoints

- POST `/v1/intent/resolve`: Describe what you want in natural language; unbrowse searches the marketplace, captures if needed, and executes the matching API call
- GET `/v1/skills`: List all skills in the marketplace with metadata, scoring, and endpoint counts
- POST `/v1/search`: Semantic search for skills by intent — returns ranked results with similarity scores
- POST `/v1/search/domain`: Search skills scoped to a specific domain
- POST `/v1/agents/register`: Register as an agent and get an API key for authenticated requests
- GET `/v1/stats/summary`: Platform-wide stats (total skills, endpoints, agents, executions)

## Use Cases

- **Web scraping without browsers**: Extract structured data from any website via direct API calls
- **Price monitoring**: Track prices on e-commerce sites by calling their internal pricing APIs
- **Travel automation**: Search flights, hotels, listings on Airbnb, Booking.com, etc. via their undocumented APIs
- **Data extraction**: Pull structured data from sites that lack official APIs
- **Agent tooling**: Give any AI agent instant access to website functionality without DOM interaction

## Integrations

Works with any tool that supports CLI execution or the OpenClaw skill protocol:

- **Claude Code**: Install via `npx unbrowse setup` — automatically detected
- **Cursor**: Install via `npx unbrowse setup` — adds MCP server configuration
- **OpenClaw**: Install via `npx skills add unbrowse-ai/unbrowse`
- **Any CLI agent**: Call `unbrowse resolve --intent "..." --url "..."` directly

## Source Code

- [GitHub Repository](https://github.com/unbrowse-ai/unbrowse): Full source code, issue tracker, and contribution guidelines (AGPL-3.0 license)
- [npm Package](https://www.npmjs.com/package/unbrowse): Published package for installation

## Full Documentation

- [skill.md](https://www.unbrowse.ai/skill.md): Complete skill documentation with all endpoints, auth flows, and rules