{'>'}_
English 繁體中文 日本語 한국어 Deutsch Español Français
Login

Capture once, replay everywhere — Unbrowse turns websites into reusable API routes for agents. Resolve an intent + URL to a ranked endpoint shortlist, execute the chosen one for real data, or open a managed browser when capture is needed. Available as an MCP server, CLI, and SDK.

702 stars
59 forks
TypeScript
145 views
View on GitHub Add to Favorites

SKILL.md

name: unbrowse description: Capture once, replay everywhere — Unbrowse turns websites into reusable API routes for agents. Resolve an intent + URL to a ranked endpoint shortlist, execute the chosen one for real data, or open a managed browser when capture is needed. Available as an MCP server, CLI, and SDK. metadata: type: integration origin: unbrowse-ai/unbrowse

Unbrowse

Unbrowse turns websites into reusable API routes for agents. Teach a route once, store sanitized metadata, replay it on later calls — typically 30× faster and 90× cheaper than a fresh browser session (peer-reviewed benchmark across 94 live domains: 3.6× mean speedup, 40× fewer tokens — arXiv:2604.00694).

This skill installs the instructions. The runtime (CLI binary + MCP server) comes from the unbrowse npm package — get it with npx -y unbrowse.

Surfaces

Surface Reach for it when
MCP server An MCP-host agent (Claude Code, Claude Desktop, Cursor, Codex, Windsurf). Tools like unbrowse_resolve, unbrowse_execute, unbrowse_go appear in the host.
CLI (unbrowse) A shell or script that wants the same surface without an MCP host.
SDK (unbrowse/sdk) A TypeScript program embedding Unbrowse; the SDK spawns its own local binary.

The workflow

Two tools, never one and never three:

  • resolve — "is there an indexed route for this intent + URL?" Returns a ranked shortlist, or a hard handoff to a browser session.
  • execute — pick one endpoint from the shortlist and run it. Returns the real data.
  • browse-session — when the API is too dynamic to predict, open a managed browser; local capture indexes the route metadata for next time.

Quickstart

Wire the MCP server into an MCP host:

{
  "mcpServers": {
    "unbrowse": {
      "command": "npx",
      "args": ["-y", "unbrowse", "mcp"]
    }
  }
}

Then once:

npx unbrowse setup

From a shell:

unbrowse resolve --intent "search hacker news for openai" --url https://news.ycombinator.com
unbrowse execute --skill-id <id-from-resolve> --endpoint-id <id-from-shortlist>

From Node:

npm i unbrowse

import { Unbrowse } from 'unbrowse/sdk';
const client = new Unbrowse({ apiKey: process.env.UNBROWSE_API_KEY });
const resolved = await client.resolve({ intent: 'search hn for openai', url: 'https://news.ycombinator.com' });
const result = await client.execute({ endpoint_id: resolved.available_operations![0].endpoint_id, params: { q: 'agents' } });

Passive parallel indexing

While you browse, Unbrowse reverse-engineers the page's API traffic and indexes it to publish — in the background, in parallel, so browsing stays fast. This is on by default; each sync checkpoint returns immediately instead of blocking on the page's network settle. Opt out per machine to index inline (slower checkpoints, complete endpoints before each returns):

unbrowse settings --passive-index off    # opt out; back on with: --passive-index on

Agents can toggle the same flag via the unbrowse_settings tool (passive_index: true|false).

Links