Skip to content

About

Local-first browser agent training: record a Chrome task once, compile parameterized workflows, run via MCP (Claude/Cursor) or export Playwright & Stagehand. Privacy-first MIT OSS—recordings never leave your machine.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

BrowserLearn

M8ven Score

Browser agent training for developers — record a real Chrome workflow once, compile it into a reusable parameterized skill, and run it from MCP clients (Claude Code, Cursor) or export Playwright / Stagehand scripts.

Free and open source (MIT). Local-first and privacy-preserving: recordings stay on your machine and are redacted before they are saved. No cloud account required.

Record Chrome MV3 extension captures clicks, types, selects, navigates
Compile Deterministic local compiler turns a recording into a workflow (optional LLM naming via BYOK / Ollama)
Run MCP tools, CLI runner, or exported automation code

Why BrowserLearn

  • Train browser agents by demonstration — no hand-written selectors for every step
  • Works with AI coding tools — MCP server for Claude, Cursor, and other Model Context Protocol hosts
  • Exports — Playwright and Stagehand for teams that want code in git
  • Private by default — typed values and recordings never leave your machine; optional LLM enrichment sends only an anonymized skeleton you can preview

Requirements

  • Node.js 22.13 or newer
  • Google Chrome (recording + runs). Without Chrome: npx playwright install chromium once for the runner

Quick start

npx browserlearn

Packages on npm: browserlearn and @browserlearn/core (0.1.0).

Extension zip (Load unpacked): v0.1.0 release.

  1. Open the dashboard URL printed in the terminal.
  2. Load the unpacked extension from the release zip (or build packages/extension/dist from git — see packages/extension/README.md).
  3. Settings → copy the pairing code into the extension popup.
  4. Open a normal website → Record → do the task → Stop.

From a git checkout instead of npm:

git clone https://github.com/droidmaximus/browserlearn.git
cd browserlearn
npm install
npm run build --workspace @browserlearn/core
npm run build --workspace browserlearn
npm run build --workspace @browserlearn/extension
node packages/cli/dist/bin.js

MCP setup, CLI, and optional LLM naming: packages/cli/README.md.

MCP (Claude Code / Cursor)

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

Tools: list_workflows, get_workflow, run_workflow, search_recordings.

Packages

Agent skills (Agensi)

Instruction pack for Claude Code, Cursor, Copilot, and other agents: install/pair, run workflows via MCP, and export Playwright/Stagehand. Source lives in skills/agensi. Free to list on Agensi; BrowserLearn itself stays MIT with no paywall.

Releases

Keywords

browser automation, browser agent, MCP, Model Context Protocol, Playwright, Stagehand, Chrome extension, AI agents, local-first, privacy, TypeScript, Claude, Cursor

License

MIT

About

Local-first browser agent training: record a Chrome task once, compile parameterized workflows, run via MCP (Claude/Cursor) or export Playwright & Stagehand. Privacy-first MIT OSS—recordings never leave your machine.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages