Search skills/

bb-browser Claude skills

A CLI and MCP server that lets agents drive your real, logged-in Chrome and run site adapters (103 commands, 36 platforms), with two small skills describing the commands.

epiral/bb-browser 2 skills Browser automation & scraping
Our verdict Recommended with caveats

Useful and open source, but by design it gives an agent and community-written adapters access to your logged-in browser sessions; use a separate Chrome profile.

3/5 Depth expertise shipped ! Security Broad access 5mo Maintenance Updated 5 mo ago 6.3K Adoption stars OK License MIT, commercial OK

What bb-browser does

bb-browser (BadBoy Browser) connects an agent to the Chrome you already use, so it can read or act on pages with your existing logins instead of a headless browser. Install it from npm; it works through a Chrome extension and daemon, or through OpenClaw's built-in browser with a --openclaw flag, or as an MCP server.

The two skills in this repo are descriptions of the CLI, not the CLI itself. bb-browser covers opening pages, snapshotting interactive elements, clicking and filling, fetch calls made inside the browser context with your cookies, network inspection and mocking, and the site system: community adapters that turn pages into commands such as twitter/search, reddit/hot, github/repo, youtube/transcript or eastmoney/stock. Four reference files cover the site system, adapter development, fetch and network, and snapshot refs. bb-browser-openclaw is a shorter version that always passes --openclaw. Adapters are one JavaScript file per command, pulled with bb-browser site update, and run via eval or fetch inside your browser tab, which is what makes them see your logged-in state. Both skills allow Bash(bb-browser:*). The last commit we saw was in May 2026.

Good for

  • Reading data from sites where you are logged in and no API exists
  • Agents that need your real browser session instead of a headless one
  • Writing new site adapters with an agent

Not for

  • Shared or sensitive browser profiles
  • Teams that forbid agents from using personal logins

What's inside depth 3/5

Two skills (11 KB and 4 KB) with four reference files and no scripts in the skill folders; the real functionality is in the separate npm CLI.

Skills
2
Reference docs
4
Scripts
0
Tests / evals
0
Templates
0
Avg SKILL.md
7.6 KB

Counted from the repository file tree. How we score depth.

Dependencies and costs

  • bb-browser npm package (Node.js 18+)
  • Chrome with the bb-browser extension, or OpenClaw's browser
  • Network access to the community bb-sites adapter repository

Security review

Broad access

This skill asks for more access than most, or sends data to another service. That is usually a design choice rather than bad intent, but you should know before you install it.

  • Uses your browser cookies or logged-in profileCore design: commands and site adapters run fetch()/eval inside your logged-in Chrome tabs using your cookies (credentials: 'include'), a cookies command lists a tab's cookies, and history-sqlite.ts copies the Chrome/Edge/Brave/Arc History database to a temp file to query it.packages/cli/src/commands/fetch.ts:65
  • Loads instructions from a server at run timebb-browser site update clones https://github.com/epiral/bb-sites.git into ~/.bb-browser/bb-sites, and after any site command a detached git pull --ff-only runs silently (line 834); those community adapters are then executed in your logged-in tabs, so behavior changes without a change to this repo.packages/cli/src/commands/site.ts:31

A browser-automation CLI and daemon that controls your real, logged-in Chrome (or an OpenClaw browser) and calls sites with your cookies; the skill pre-approves all bb-browser shell commands. Community site adapters are pulled from a separate GitHub repo and updated silently, then run inside your signed-in tabs. A local daemon on 127.0.0.1 can optionally register with a remote hub.

Worth knowing

  • Runs remote code at an unpinned versionRuns npx openclaw ... without a version for the OpenClaw browser bridge, and site.ts calls npm view bb-browser version for update checks.packages/cli/src/cdp-discovery.ts:35
  • Makes network requestsOptional daemon start --hub https://hub.pinixai.com --hub-token ... registers the browser with a remote Pinix Hub (WebRTC video and TURN credentials from the hub); the Dockerfile also downloads Chrome and a viewer binary with curl.packages/daemon/src/hub-bridge.ts:1
  • Installs packages or toolsDocker image setup downloads bb-viewer and Chrome for Testing with curl at build time.Dockerfile:26
  • Relies on a CLI or package we could not reviewShells out to sqlite3, git, npm and npx; the compiled CLI and extension are published to npm.packages/cli/src/history-sqlite.ts:73

What we checked

The repository (105 files) on Oct 29, 2026. Not a full review: we read 2 of 73 code files; it relies on code outside the repository: npm package bb-browser and the Chrome extension it relies on (published build); Community adapter repository github.com/epiral/bb-sites (cloned and updated at runtime) and Pinix Hub service; Docker-downloaded bb-viewer binary and Chrome for Testing.

NVIDIA SkillSpector static scan: 70 pattern hits (Supply Chain 42, Data Exfiltration 10, Tool Misuse 6, Server-Side Request Forgery 5, Prompt Injection 2, MCP Rug Pull 2). The 'keychain' hit is a --use-mock-keychain Chrome flag for the managed browser; 'Supply Chain' LOWs are caret version ranges in package.json; 'rm' hits are Dockerfile cleanup. The Hidden Instructions hits are HTML comments citing source lines in a reference document. PRIVACY.md states no telemetry and local-only operation, which matches what the grep showed apart from the optional hub mode.

A review, not a certification. Skills change after we check them — read the files before installing.

License

MIT — commercial use allowed.

Can it run unattended?

Not really. It acts through your live browser session and expects you to be logged in; the skills have no stop conditions or verification loop.

Alternatives to bb-browser

agent-browser

vercel-labs/agent-browser

A native browser-automation CLI for agents, plus 11 skills that teach the snapshot-and-ref workflow for navigating, filling forms, scraping and testing.

Recommended Check first Depth 4/5

Browser Use

browser-use/browser-use

Seven skills around the Browser Use browser agent: drive local Chrome over CDP, use the paid cloud API, run QA scoring, or pay by crypto wallet.

With caveats Broad access Depth 3/5

BrowserAct Skills

browser-act/skills

A browser-automation CLI skill with stealth browsers, CAPTCHA solving and handoff to a human, plus about 100 ready-made scraper skills for shops and lead sources.

With caveats Broad access Depth 3/5

playwright-cli

microsoft/playwright-cli

Microsoft's command-line interface for driving a browser through Playwright, plus a skill that teaches coding agents the commands.

Recommended Check first Depth 3/5

FAQ

What is bb-browser?

It is an open-source CLI and MCP server that lets AI agents use your real Chrome with your logins. It ships 103 adapter commands across 36 platforms, such as Reddit, GitHub, YouTube and Zhihu, plus general click, fill and fetch commands.

How do I install bb-browser?

Run npm install -g bb-browser, then bb-browser site update to pull adapters. For MCP, add a server entry that runs npx -y bb-browser --mcp. OpenClaw users can pass --openclaw and skip the Chrome extension.

Is bb-browser safe?

It works by running code in your logged-in browser tabs, so a bad adapter or a prompt injection could act as you on those sites. Use a dedicated Chrome profile, review adapters, and avoid it on accounts with payment or admin rights.

bb-browser vs agent-browser?

agent-browser is a general automation CLI with its own browser. bb-browser reuses your existing logged-in Chrome and adds ready-made site commands. Choose bb-browser when you need your sessions, agent-browser for isolated automation.

Sources: GitHub repository (stars, license, file tree, last push May 29, 2026). Review based on the README and SKILL.md files as published.