Agent Reach
Set up and diagnose multi-platform internet access for shell-capable AI assistants.
- Source repo
- Panniantong/Agent-Reach
- Stars
- ★ 109k
- Last updated
- 1d ago
- License
- MIT
- Primary language
- Python
- FA score
- 62/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformClaude Code
- Cost
- Free, no paid service needed
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A Claude Code developer wants quick access to public GitHub repositories and search, then authenticated access to private repositories, issues, pull requests, or forks.
- Not a fit if
- Teams that need fully offline operation without platform logins
- Users unwilling to risk account bans or use a burner account
- Developers wanting one unified reading API instead of upstream tools
- Source review
- 62/100 · Some gaps
What does this agent do, and when should you use it?
Agent Reach is a Python 3.10+ command-line capability layer that installs, configures, and diagnoses internet-access tooling for AI assistants that can run shell commands. Rather than wrapping platform data access behind its own API, it selects usable backends from ordered channel routes and has the agent invoke upstream tools directly. Its documented channels include web pages, YouTube, RSS, GitHub, X, Reddit, Bilibili, Xiaohongshu, Facebook, Instagram, LinkedIn, V2EX, Xueqiu, and web-scale semantic search. Installation can configure Exa through MCP, register a SKILL.md, and use agent-reach doctor to report each channel’s active backend and remediation guidance. It fits users who want one command-line agent environment for public web content, video transcripts, code-hosting content, and selected login-gated social platforms.
When agent-reach install runs, the project installs the agent-reach CLI and detects or installs Node.js, gh CLI, and mcporter; the README also says it connects Exa through mcporter MCP without an API key and registers a SKILL.md in the agent’s skills directory. channels/web.py uses Jina Reader for web pages, youtube.py uses yt-dlp for transcripts, rss.py uses feedparser for RSS/Atom, and github.py uses gh CLI for GitHub reading and search. doctor actively probes channel candidates in order, including twitter-cli, OpenCLI, rdt-cli, bili-cli, xiaohongshu-mcp, and linkedin-scraper-mcp, then reports the selected backend or a repair prescription. Search, reading, and extraction are performed by the agent calling those upstream tools directly, not by a unified Agent Reach data API.
- A Claude Code developer wants quick access to public GitHub repositories and search, then authenticated access to private repositories, issues, pull requests, or forks.
- A researcher needs to extract YouTube subtitles, search videos, or turn ordinary web pages into readable content with Jina Reader.
- An operator working on a desktop with an existing login session needs to search or read X, Reddit, Facebook, Instagram, or Xiaohongshu.
- A user tracking RSS or Atom sources wants an agent to read feeds through feedparser.
- A team deploying command-line agents on workstations or servers needs agent-reach doctor to check channel availability and backend selection.
How do you install or deploy this agent?
The documented runtime is Python 3.10+, and the agent needs permission to execute shell commands. Run: agent-reach install --env=auto. For safe mode, run: agent-reach install --env=auto --safe. To preview actions without changes, run: agent-reach install --env=auto --dry-run. The README also documents giving this instruction to Claude Code, OpenClaw, Cursor, or similar agents: “帮我安装 Agent Reach：https://raw.githubusercontent.com/Panniantong/agent-reach/main/docs/install.md”. OpenClaw users must enable exec permissions first, for example with openclaw config set tools.profile "coding", then restart the Gateway and begin a new conversation.
How do you use this agent?
After installation, run agent-reach doctor to inspect channel status, the active backend, and repair guidance. The documented no-extra-configuration paths include Jina Reader for web pages, yt-dlp for YouTube subtitles, bili search for Bilibili, gh repo view owner/repo for public GitHub repositories, and feedparser for RSS. For login-dependent services, ask the agent to configure the relevant platform, such as Twitter, Reddit, or Xiaohongshu. X cookies must be manually exported by the user with Cookie-Editor; direct use of the upstream twitter command also requires TWITTER_AUTH_TOKEN and TWITTER_CT0 to be explicitly set in the current process environment.
What are this agent's strengths and limitations?
- It models each channel as an ordered primary-and-fallback route and has agent-reach doctor actively test candidates instead of merely checking whether a command exists.
- It brings web reading, YouTube transcripts, RSS, GitHub, and Exa semantic search into one installation and diagnostic workflow for command-line agents.
- It documents distinct paths for platforms requiring login state, including desktop Chrome sessions, manually exported Cookie-Editor cookies, and server-oriented backends.
- It is a capability layer, not a unified reading API: the agent still invokes upstream tools such as Jina Reader, gh, yt-dlp, and OpenCLI.
- X, Reddit, Xiaohongshu, Facebook, and Instagram depend on user login state or cookies, and the README warns that scripted or API access can risk account restrictions or bans.
- Adoption can require Python, Node.js, gh CLI, mcporter, and channel-specific upstream tools; server deployments may also need a proxy.
- Some channels have no zero-configuration route; the README explicitly states that Reddit’s anonymous interface is blocked and a logged-in path is required.
How does this agent compare with similar options?
Compared with installing Jina Reader, yt-dlp, gh CLI, feedparser, twitter-cli, or OpenCLI separately, Agent Reach is positioned as the layer that selects, installs, health-checks, and switches among those upstream tools. It does not replace them for the underlying reads.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Agent Reach This agent | 62 · Some gaps | CLIFree | ★ 109k | 1d ago | Python | Claude Code |
| CompozyOS | 52 · Major gaps | CLIFree + model costs | ★ 2.8k | 1d ago | Go | ChatGPT · Codex · Claude Code |
| Native Search Hub | 24 · Major gaps | CLIFree | ★ 1.3k | 6mo ago | Python | Codex · Claude Code |
| OWL Multi-Agent Automation | 39 · Major gaps | CLIFree + model costs | ★ 20k | 4d ago | Python | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows default install does not modify system, requires explicit --system; cookies stored locally with 600 permissions; dry-run and uninstall available. But dependency security only claims open source, no audit or version pinning; external effects have account ban risk warning but no mitigation.
Docs and tests consistent, channel registry has contract tests; but dependency availability relies on external tools (yt-dlp, bili-cli) without version pinning or mirrors; failure messages have doctor diagnostics but no specific error handling examples.
Targets multiple agents and platforms, scenarios clear; capability boundaries clear (zero-config vs config required); trigger words explicit (install/update/configure); environment fit has local/server distinction but no detailed Windows/macOS instructions.
Information architecture clear with README, docs, guides; install notes detailed; naming stable (agent-reach); examples and FAQ present; known limitations documented; MIT license; version number exists but no changelog; maintenance responsibility declared but no contribution guide.
Output usability high, provides structured results; marginal value high, solves Agent internet access pain point; cost-benefit good, free and open source, but external services may incur costs.
Claims supported by tests, but no test results provided; cross-source corroboration insufficient, relies on single repo; fact/inference separation, docs distinguish current selection and design philosophy.
- External tools (yt-dlp, bili-cli) may become unavailable, requiring regular updates;
- Cookie usage carries account ban risk, recommend using secondary accounts;
- No dependency version pinning, supply chain security needs attention.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
中文： - 全网搜索、Reddit、小红书、B站在 1.6.0 里会出现的几种"明明能查却报错"或"显示成功但内容是空的"，这版都修好了。 - 电脑上别的软件占用了浏览器端口时，可以换一个端口，不用关掉那个软件。 - 复制登录失败时会如实告诉你，不再假装成功。 - Windows 上小红书带特殊符号的链接能正常打开了。 English: - Several 1.6.0 cases where web search, Reddit, Xiaohongshu or Bilibili failed although the content was reachable, or reported success with empty content, are fixed. - When another program on your computer uses the browser port, Agent Reach can use a different one; you don't have to close that program. - A login copy that fails now says so instead of reporting success. …View on GitHub ↗ -
Release notes excerpt
中文： - 从旧版升级：把"帮我更新 Agent Reach"那句话发给你的 AI 就行。你的密钥和设置都保留，旧版程序会自动删掉。 - 加一个平台只要说一句，需要的软件都会自动装好，不需要管理员密码；你只需要登录或者填一个 Key。 - 小红书、推特这类要登录的平台，改用 Agent Reach 自己的专用浏览器，不弹窗、不占用你的浏览器；多个 AI 可以同时查。 - RSS 订阅和 Exa 全网搜索合并进了"网页"。以后有新版本，AI 会用大白话告诉你改了什么，你同意了才更新。 English: - Upgrading from an older version: just send your AI the "update Agent Reach" sentence. Your keys and settings are kept, and the old program is removed for you. - Adding a platform takes one sentence. Everything it needs is installed for you, with no admin password; you only log in or enter a key. …View on GitHub ↗ -
v1.5.0 v1.5.0 — 能力层:多后端路由 + 真体检 + OpenCLI
Release notes excerpt
这一版是什么 Agent Reach 从「工具集合」升级为「能力层」:每个平台 = 首选 + 备选的有序后端列表,当下最稳的接入方式替你实测选好、装好、体检好——接入方式换代,你无感。本版全部选型基于 2026-06 真机实测(32 项端到端测试)。 平台路由变化 | 平台 | 现在的路 | |------|---------| | 小红书 | OpenCLI(桌面,复用浏览器登录态零配置)▸ xiaohongshu-mcp(服务器,扫码)▸ xhs-cli(存量备选,上游已停更) | | Reddit | OpenCLI ▸ rdt-cli。诚实口径:Reddit 没有零配置路径(匿名接口已被全面封锁、官方 API 审批制),必须登录态 | | B站 | bili-cli(搜索/热门/详情,无需登录)▸ OpenCLI(字幕)▸ 搜索 API 兜底。yt-dlp 退出 B站(被 412 风控全面拦截,实测最新版+代理+Cookie 均无效);yt-dlp 继续服务 YouTube | | Twitter | twitter-cli 守门(实测稳定)▸ OpenCLI 备选,探测加重试 | doctor 成了真体检 …View on GitHub ↗