What is Agent Reach?
Agent Reach is a command-line tool plus an agent skill that gives Claude Code, Codex, OpenClaw and similar agents one command, agent-reach, for searching and reading platforms their built-in web tools handle badly. Default channels cover Bilibili, V2EX, GitHub, YouTube and general web pages; optional ones add XiaoHongShu, Xueqiu, Boss Zhipin, Xiaoyuzhou podcasts, LinkedIn jobs and Reddit; X, Facebook and Instagram are marked experimental. It is mostly glue: each action tries an ordered list of upstream tools such as OpenCLI, bili-cli, yt-dlp, Jina Reader and Exa, falling back when one breaks.
For sites that need a login it runs its own headless browser, preferably CloakBrowser, a stealth Chromium build, and, if your everyday Chrome has the OpenCLI extension connected, copies that one site's cookies from it the first time the site is used. It is read-only: it does not post, comment, like or message.
The code is MIT, written mostly by one maintainer. Version 1.6.0, a rebuild, shipped on 23 October 2026. The README is in Chinese with an English translation, and warns that the PyPI package named agent-reach is a different project.
Is Agent Reach worth it?
The 'zero API fees' line is accurate and is also the problem. Agent Reach avoids paying X, Reddit or LinkedIn by driving a logged-in browser through their websites, and X's and LinkedIn's terms forbid exactly that. X's terms say scraping in any form without written consent is 'expressly prohibited' and set liquidated damages of $15,000 per 1,000,000 posts for anyone viewing more than a million in 24 hours. LinkedIn's user agreement bans scraping and bots. The README itself advises using a secondary account because platforms may restrict accounts used for automated access. That is the honest framing: it works, until the platform notices.
The account risk is concrete because of how logins work. On first use of a site, Agent Reach copies that site's cookies from your Chrome into its own profile under ~/.agent-reach/, through the OpenCLI browser extension. It does not read your password store, but the copied session is usually your main account, and it persists until you log out of the channel. The dedicated browser listens only on your machine, and the tool rate-limits itself to 2 concurrent queries per platform, 4 overall and a 10-second gap between searches on one platform.
The install path deserves the same scrutiny. The recommended method is to paste a raw GitHub URL into your agent and let it follow the instructions, which run uv tool install against the latest release with no pinned version. Channel setup then installs Node, OpenCLI, a browser extension and the CloakBrowser binary, which is proprietary and requires a paid subscription for its latest major version. An unrelated PyPI package of the same name exists, and the README carries a separate warning about crypto tokens using the project's name.
The engineering is more careful than the pitch. The 1.6.0 changelog reports that 28 of 40 actions passed real acceptance tests before release, and lists the ones that shipped untested: Reddit, Instagram, Bilibili subtitles, LinkedIn, Xueqiu hot lists and Xiaoyuzhou. That transparency is welcome. For GitHub, YouTube subtitles and public web pages, the official APIs and tools it wraps are just as reachable directly; its distinctive value is the Chinese platforms, where general search coverage is poor.
Where it earns its place
A market analyst covering Chinese consumer brands
Have Claude Code summarise XiaoHongShu notes and Bilibili videos about a product launch
Add the xiaohongshu channel, sign in with a secondary account in the dedicated window, and ask the agent to search and read notes and subtitles. Cost is model tokens; the risk is the account being restricted.
A developer researching a library before adopting it
Pull GitHub issues, V2EX threads and YouTube talk transcripts into one briefing
The default github, v2ex and youtube channels work without a login; transcription of videos without subtitles needs a Groq or OpenAI key at that provider's rates.
A recruiter at a 20-person firm in Shenzhen
Search Boss Zhipin and public LinkedIn job posts for salary benchmarks
Public LinkedIn jobs need no login, but LinkedIn's user agreement still bans scraping and bots, and the changelog says the LinkedIn channel shipped untested. Keep volumes small and treat results as a sample.
Before you commit
- Setup
- Send your agent the install.md URL, or run uv tool install against the GitHub release zip followed by agent-reach setup. Optional channels are added with agent-reach add <channel>, which installs Node, OpenCLI, a browser extension and CloakBrowser, falling back to installed Chrome.
- Integrations
- Works with any agent that can run shell commands; the skill is documented for Claude Code, Codex, Hermes Agent and OpenClaw. Every command returns one JSON object. Upstream tools include OpenCLI, bili-cli, boss-agent-cli, yt-dlp, Jina Reader and Exa.
- Data and privacy
- Everything lives under ~/.agent-reach/, including copied site cookies and saved sessions. Page requests go to the platforms, Jina Reader and Exa; transcription audio goes to Groq or OpenAI. No published security advisories; SECURITY.md promises acknowledgement within 48 hours.
- Cost at scale
- Free apart from transcription keys and, for the newest CloakBrowser, a CloakHQ subscription; the README's sponsors are proxy and scraping vendors. The real cost at volume is legal: X's terms set $15,000 per million posts over a million a day.
- Limits
- About 246 open issues. X, Facebook and Instagram are experimental and may break at any time; Xueqiu sessions expire within hours. Queues wait up to 2 minutes before returning BUSY. OpenCLI is pinned to 1.8.7 because 1.8.8 broke XiaoHongShu.
What's good
- One command for Bilibili, XiaoHongShu, V2EX, Xueqiu and Boss Zhipin
- Read-only by design; never posts or messages
- Ordered fallbacks when an upstream tool breaks
- Changelog states which actions shipped untested
What's not
- Automated access to X and LinkedIn breaches their terms
- Can copy your main-account cookies from Chrome
- Unpinned install driven by an agent reading a remote URL
- Depends on a proprietary stealth browser binary
Best for
Individual researchers and developers who need their agent to read Chinese platforms and accept the account and terms-of-service risk on a secondary account.
Agent Reach alternatives
Agent Reach: common questions
- Is Agent Reach worth it?
- Agent Reach is a clever convenience for personal research on Chinese and social platforms, and a liability anywhere an employer, a platform's terms or your main account matter. We rate it 3.2 out of 5.
- How much does Agent Reach cost?
- Free under MIT. Some channels need your own keys: YouTube transcription uses a Groq or OpenAI key, podcast transcription a Groq key. The newest CloakBrowser binary needs a paid CloakHQ subscription.
- What are the downsides of Agent Reach?
- Automated access to X and LinkedIn breaches their terms; can copy your main-account cookies from Chrome; unpinned install driven by an agent reading a remote URL; depends on a proprietary stealth browser binary.
- Who is Agent Reach best for?
- Individual researchers and developers who need their agent to read Chinese platforms and accept the account and terms-of-service risk on a secondary account.
- How much does Agent Reach cost for a team or in production?
- Free apart from transcription keys and, for the newest CloakBrowser, a CloakHQ subscription; the README's sponsors are proxy and scraping vendors. The real cost at volume is legal: X's terms set $15,000 per million posts over a million a day.
- Is Agent Reach safe for company data?
- Everything lives under ~/.agent-reach/, including copied site cookies and saved sessions. Page requests go to the platforms, Jina Reader and Exa; transcription audio goes to Groq or OpenAI. No published security advisories; SECURITY.md promises acknowledgement within 48 hours.
- What are the best Agent Reach alternatives?
- The closest alternatives we have reviewed are Crawl4AI, Firecrawl, Browser Use and notebooklm-py.
Answers checked Oct 26, 2026.