Agent Reach gives an AI agent read and search access to the web and social platforms that are hard to reach from a command line: paid APIs, IP blocks, login walls, and raw HTML. You hand your agent a one-line install instruction, and the agent sets up a CLI plus a SKILL.md file that tells it which upstream tool to call for each platform. Any agent that can run shell commands can use it, including Claude Code, OpenClaw, Cursor, and Windsurf.
It describes itself as a capability layer rather than another scraper. It selects, installs, health-checks, and routes to upstream tools, and the agent then calls those tools directly with no wrapper in between. Each platform is an ordered list of a preferred backend and fallbacks, probed for real availability, so a broken backend is swapped by reordering the list. A doctor command reports which backend each channel is using and how to fix the ones that fail.
Supported channels and what they unlock:
| Platform | Works out of the box | After configuration |
|---|---|---|
| Web pages | Read any page through Jina Reader | |
| YouTube | Subtitle extraction and video search | |
| RSS | Read any RSS or Atom feed | |
| Web search | Semantic search through Exa over MCP | |
| GitHub | Read and search public repositories | Private repos, issues, pull requests, forks |
| Twitter/X | Read single posts | Search, timelines, long-form posts |
| Bilibili | Search and video details | Subtitles |
| Search, posts, and comments | ||
| Facebook, Instagram | Search, profiles, feeds | |
| Xiaohongshu | Search, reading, comments | |
| Public pages | Profiles, company pages, job search | |
| V2EX, Xueqiu | Posts, nodes, stock quotes | |
| Xiaoyuzhou podcasts | Audio transcription with Whisper |
Cookies for logged-in platforms stay on the local machine. The default install only checks the environment, and it installs system packages or writes agent configuration only when you explicitly allow it.
Features
- One-line setup: an agent reads an install instruction and configures the CLI, upstream tools, and skill file itself
- Fallback routing: each platform has a preferred backend and ordered fallbacks, switched without code changes
- Health checks: a doctor command shows each channel's status, active backend, and fix for anything broken
- Zero-config channels: web pages, YouTube subtitles, RSS, GitHub, Bilibili search, and V2EX work without login
- Login-gated channels: Twitter/X, Reddit, Facebook, Instagram, Xiaohongshu, and LinkedIn unlock with the user's own browser session or cookies
- Semantic web search: Exa search connected over MCP without an API key
- Agent-agnostic: works with any agent that runs shell commands, guided by a SKILL.md file
- Local credentials: cookies are stored only on the local machine, and system changes require explicit permission
