bb-browser is positioned to let AI agents access the internet using the human interface rather than APIs. The project points out that 99% of websites do not provide APIs, and the internet was built for browsers in the first place. bb-browser's approach is: rather than forcing websites to provide machine interfaces, let machines use the human interface directly. Its adapters run eval in your browser tab, call fetch() with your cookies, or invoke the page's own webpack modules—the website believes the request comes from you, because in fact it does.
Main capabilities and coverage:
- Directly reuses your real browser's login state, with no keys, no bot accounts, and no scrapers.
- Provides 36 platforms and 103 commands, with example platforms including Twitter, Reddit, YouTube, Zhihu, Bilibili, LinkedIn, and GitHub.
- Supports OpenClaw: run directly through OpenClaw's built-in browser, with no Chrome extension or daemon required. The skill on ClawHub is named bb-browser-openclaw.
- Supports MCP (Claude Code / Cursor) integration.
- It is also a complete browser automation tool: all commands support --json output, --jq <expr> inline filtering, and --tab <id> concurrent multi-tab operations.
How to use it:
- It can be used directly through OpenClaw's built-in browser, with no extension required; it can also be integrated as an MCP server into Claude Code or Cursor.
- The daemon binds to 127.0.0.1:19824 by default, and --host can be used to customize the host.
- The community is driven through bb-sites, with each command being a JS file; adding any website takes about 10 minutes. You can have an AI agent execute "turn the XX website into a CLI," and it will read the guide, reverse-engineer the API with network --with-body, write the adapter, test it, and submit a PR, all autonomously.
- Adapter complexity is divided into three levels.
Applicable scenarios:
- Have AI agents perform actions or scrape structured data on websites without official APIs.
- Conduct cross-platform research: the material states that an agent can, in less than a minute, across six platforms and six dimensions, obtain structured JSON, with speed and breadth exceeding any human researcher.
- Expand the scope of the internet accessible to agents: without bb-browser, an agent's world is files, terminals, and a small number of APIs with keys; with it, it is files, terminals, and the entire internet.
Limitations and caveats:
- It depends on your real browser's login state, so its capability boundaries are related to login state.
- Regarding the complete list of specifically supported commands, the specific content of the adapter's three levels of complexity, installation commands, dependency requirements, license agreement, and pricing information, the material does not currently provide them.
Philosophy and validation: The project emphasizes "Your browser is the API. No keys. No bots. No scrapers." and states that it has been validated: 20 AI agents running in parallel, each independently reverse-engineering a website and producing usable adapters, and the marginal cost of connecting a new website to the internet accessible to agents is approaching zero.
