Panniantong/Agent-Reach

Agent Reach – One‑click Internet Access for Your AI Agent

What it is – A Python‑based command‑line utility that equips any LLM‑powered “agent” (Claude Code, OpenClaw, Cursor, Windsurf, etc.) with ready‑made, up‑to‑date ways to read web pages, fetch YouTube subtitles, query Reddit, pull GitHub data, scrape Twitter/X, and many other popular platforms. It does not implement the crawling or parsing itself; instead it selects, installs, and health‑checks the best open‑source tools for each service and wires them into the agent via a simple skill file.

Why you might need it – Modern agents can write code or manage projects, but they stumble when asked to “look that up on the web”. Each site usually requires a different CLI tool, API key, login cookie, or a fragile workaround. Agent Reach abstracts that plumbing away: a single natural‑language command like

帮我安装 Agent Reach：https://raw.githubusercontent.com/Panniantong/agent-reach/main/docs/install.md

makes the agent automatically install the required binaries, configure credentials (if you provide cookies), and register a SKILL.md so the agent knows which command to call for a given request.

Key features (as described in the README)

• Zero‑config channels for plain web pages, YouTube subtitles, RSS feeds, and public GitHub repos – just works out of the box.
• Optional login‑required channels (Twitter/X, Reddit, Facebook, Instagram, 小红书, LinkedIn, etc.) – the agent can guide you through cookie export or login‑state setup.
• Automatic backend selection – each platform has a prioritized list of tools (e.g., twitter-cli → OpenCLI → bird). The first fully functional backend is used; if it breaks, agent‑reach doctor shows the fallback.
• Self‑diagnosing – agent-reach doctor reports which channels are healthy, which backend is active, and how to fix problems.
• Safety‑first install – default install only checks the environment; real system changes happen only with --system.
• Cross‑agent compatibility – works with any agent that can run shell commands, not tied to a specific LLM.
• Open‑source and free – MIT‑licensed, all underlying tools are open source; the only possible cost is a cheap proxy server if you run the agent on a remote machine.

Supported platforms (quick view)

Platform	Ready‑to‑use	Extra setup
Web pages	✅	–
YouTube (subtitles)	✅	–
RSS/Atom	✅	–
GitHub (public)	✅	–
Twitter/X (single tweet)	✅	Export cookies
Twitter/X (search, timeline)	❌	agent-reach configure twitter
Reddit	❌	Login via OpenCLI or rdt-cli
Facebook / Instagram	❌	Login via OpenCLI (Chrome session)
Bilibili	✅	–
LinkedIn	✅ (public)	agent-reach configure linkedin for full profile access
小红书, 小宇宙播客, 雪球, V2EX, etc.	✅ (basic)	Optional credential steps

How it works

1. Installation – pip install -e . (or the provided installer script) adds the agent-reach CLI.
2. Dependency check – the tool verifies Node, gh CLI, mcporter, etc., and suggests missing installs.
3. Channel registration – Python modules under channels/ declare the preferred and fallback back‑ends for each service.
4. Skill generation – when you run agent-reach install, a SKILL.md is written into the target agent’s skill directory, mapping natural‑language intents to the appropriate shell command.
5. Runtime – the agent reads SKILL.md, decides which command to invoke (e.g., yt-dlp for YouTube subtitles), runs it, and returns the structured result to the LLM.

Security & privacy

• Cookies/tokens are stored locally in ~/.agent-reach/config.yaml with 600 permissions; nothing is uploaded.
• The installer never modifies system files unless you pass --system.
• A dry‑run mode (--dry-run) shows all actions before they happen.
• The README warns that using login cookies can risk account bans; it recommends dedicated throwaway accounts.

Typical workflow

User: 帮我看看这个链接
Agent: (reads SKILL.md) → runs `curl https://r.jina.ai/URL`

User: 帮我搜索 Reddit 上关于 LLM 的讨论
Agent: asks you to configure Reddit → you run `agent-reach configure reddit`
Agent: then runs `opencli` with your browser session and returns posts.

Who might use it

• Developers building custom AI assistants that need live web data.
• Teams deploying agents on internal servers who want a single, maintainable way to keep web‑access capabilities up‑to‑date.
• Hobbyists who want their personal Claude/ChatGPT‑style bots to browse the internet without writing scrapers.

Where to start

1. Clone the repo.
2. Run agent-reach install --env=auto (adds the CLI and checks dependencies).
3. Copy the one‑line install command into your agent’s chat.
4. Use agent-reach doctor to verify each channel.

---

Bottom line – Agent Reach is a practical, open‑source capability layer that removes the tedious plumbing required for LLM agents to fetch real‑world data. It stays in the AI‑agent ecosystem, not an unrelated tutorial or list, so it fits squarely within the scope of frontier AI tooling.