Agent Reach: one-command internet access layer for CLI agents (Xiaohongshu, Bilibili, YouTube, Xueqiu, V2EX)

What they did

A developer running a CLI agent (Claude Code / Codex / OpenClaw style) hit the usual wall: built-in search and page reading can't get past login walls, can't read video or podcast audio, and covers Chinese platforms poorly. Instead of hand-wiring a scraper per site, they installed Agent Reach, which bundles and pre-tests the tools behind each platform. They ran `agent-reach channels` to list available channels, then invoked individual channels to pull structured data: Xiaohongshu notes, X account activity, Bilibili and YouTube subtitles, YouTube and Xiaoyuzhou podcast speech-to-text, Xueqiu quotes, V2EX threads and replies, and GitHub repos and issues. The agent then reasoned over that clean structured output rather than raw HTML, so research and content tasks could cite real platform data.

How it went

Working setup: the agent can now reach platforms it previously couldn't, returning structured content instead of scraped HTML. Channel listing and per-platform calls are one command each, and the bundled tools are tested by the project, so setup is mostly install-and-go. Quality depends on the underlying platform tools and on the user's own logged-in sessions for gated features.

Worth knowing

Login-walled channels need valid authenticated sessions (cookies or browser login) that expire and must be refreshed. Scraping-heavy channels can break when platforms change, and rate limits or anti-bot measures may throttle volume. Chinese-platform coverage is the point but also the fragility. Respect each platform's terms and local law; the project is MIT-licensed but third-party tools and sponsors have their own terms.

Wiring

Install Agent Reach (Python 3.10+) and confirm the CLI is on PATH. Run `agent-reach channels` to enumerate channels and their status. For login-gated channels (Xiaohongshu, X), complete the browser login or supply cookies once so the tool can persist a session. Invoke a channel per task (e.g. subtitles for Bilibili/YouTube, speech-to-text for YouTube/Xiaoyuzhou, quotes for Xueqiu, threads for V2EX, repos/issues for GitHub), and have the agent consume the structured output. Optionally pair with OpenCLI for shell-level orchestration, and schedule a periodic credential-refresh check.

Try it yourself

Install Agent Reach in this repo's environment (Python 3.10+), then run `agent-reach channels` and show me the available channels. Pick the channels relevant to my task, verify each with a small test call, and write a short report of which channels work, which need login, and how to refresh credentials. Then use the working channels to gather the data I ask for and return it as structured JSON.