Cross-platform reading and a unified CLI
Login walls, paid APIs, and platform-specific anti-scraping mechanisms prevent AI agents from browsing Twitter threads, viewing YouTube subtitles, or reading Xiaohongshu posts independently. Agent-Reach combines community-maintained free scraping and reading tools, including twitter-cli, yt-dlp, and Jina Reader, into a unified CLI. Agents can use the same commands to read more than ten platforms with no API fees.
Supported platforms include Twitter/X, Reddit, YouTube, GitHub, Bilibili, Xiaohongshu, WeChat, Weibo, V2EX, Xueqiu, LinkedIn, TikTok, and RSS. The underlying layer wraps existing open-source tools: Jina Reader for web pages (no key needed), yt-dlp for video subtitles (154K stars), Exa via mcporter for search, and the official gh CLI for GitHub. Agent-Reach provides the interface and channel management; these upstream tools perform the actual retrieval.
The architecture uses pluggable channels, with each platform isolated in channels/PLATFORM.py; some platforms connect through MCP (Model Context Protocol). Configuration is stored locally in ~/.agent-reach/config.yaml with 600 permissions. The license is MIT, and the latest version is v1.4.0 (2026, month 3).
Installation
Login cookies
doctor check
Agent calls a channel
Content retrieval
Installation performed by the agent
Paste the official installation document URL into a supported agent, such as Claude Code or Cursor. The agent reads the document and installs according to its environment. Paste the following into the chat:
# Paste this into the agent. It will read the documentation and complete installation
Install Agent Reach for me:
https://raw.githubusercontent.com/Panniantong/agent-reach/main/docs/install.mdAfter installation, run the diagnostic command to check which channels are ready and which still need login:
# Check the status of all channels
agent-reach doctor
# Install/repair channels (use --safe or --dry-run for a preview)
agent-reach install --env=auto
agent-reach install --env=auto --dry-runSupported platforms and underlying tools
Each platform is an independent channel backed by an established open-source tool. The cards list the main channels and their underlying tools. Platforms marked cookie require login and cookie export first; those marked MCP connect through Model Context Protocol.
No login · 01
Web
Web pages Jina Reader
Converts any URL into clean, readable text without an API key. The default channel for agents reading ordinary web pages and articles.
cookie · 02
Twitter / X
Tweets twitter-cli
Reads and searches tweets and threads. Uses cookie login and twitter-cli (about 2.1K stars).
cookie · 03
Reddit
Discussions rdt-cli
Reads subreddits and discussion threads. Run rdt login first to complete cookie login.
No login · 04
YouTube / Bilibili
Video subtitles yt-dlp
Extracts video subtitles and information through yt-dlp (about 154K stars), turning long videos into readable text.
MCP · 05
Search
Semantic search Exa
Connects Exa through mcporter for AI semantic search. The free allowance needs no key and lets agents find information proactively.
Official tool · 06
GitHub
Code gh CLI
Reads repositories, issues, PRs, and code through GitHub's official gh CLI.
cookie · 07
小紅書 XiaoHongShu
Image and text posts xhs-cli
Reads Xiaohongshu posts. Requires xhs login and uses xhs-cli (about 1.5K stars).
MCP · 08
TikTok / LinkedIn
Social platforms MCP server
TikTok uses douyin-mcp-server (no login); LinkedIn uses linkedin-mcp-server (browser automation).
No login · 09
RSS
Feeds feedparser
Parses RSS/Atom feeds through feedparser (about 2.3K stars), suitable for tracking blogs and news.
Channel login requirements
Channels requiring no login work after installation. Cookie channels require browser login and cookie export; MCP channels connect through mcporter. Use agent-reach doctor to check status at any time.
| Type | Platforms | Underlying tools |
|---|---|---|
| No login | Web / YouTube / Bilibili / RSS | Jina Reader · yt-dlp · feedparser |
| Cookie login | Twitter / Reddit / Xiaohongshu | twitter-cli · rdt-cli · xhs-cli |
| MCP connection | Search / TikTok / LinkedIn | Exa · douyin-mcp · linkedin-mcp |
| Official tool | GitHub | gh CLI |
Channel login configuration
These points come from Agent-Reach's official README and documentation. Configure cookies for channels that require login and use doctor to diagnose problems. Check the latest GitHub repository documentation for details.
Channel diagnostics with doctor
Run agent-reach doctor first after installation. It reports each channel's status, identifying platforms that are ready and those missing login or tools, faster than testing each one individually.
Source · Official README · Diagnostics
Initial use of channels requiring no login
Web (Jina Reader), YouTube/Bilibili (yt-dlp), and RSS (feedparser) need no login and work after installation. Test the workflow with these first, then configure platforms that require cookies.
Source · Official README · Tool table
Cookie export with a browser extension
Twitter, Reddit, and Xiaohongshu require login. The official recommendation is to export cookies after login with Chrome's Cookie-Editor extension, then enter them as documented. This is more reliable than capturing headers manually.
Source · Official README · Cookie configuration
Platform-specific login commands
Reddit uses rdt login and Xiaohongshu uses xhs login for their respective login procedures. These commands belong to the underlying tools; follow the documentation once to complete setup.
Source · Official README · Tool table
Optional proxies for Bilibili access abroad
Bilibili access from overseas servers may be restricted by region. The official documentation mentions an optional proxy costing about 1 US dollar per month. Skip nonessential platforms initially and add access when needed.
Source · Official README · Configuration
Exa's free search allowance
Use the built-in Exa semantic search through mcporter to let agents find information proactively. The free allowance needs no key. Validate your needs on the free tier before deciding whether to upgrade.
Source · Official README · Search
Cross-platform comparison of tweets and video
This example conversation asks an agent to summarize an X thread and add a subtitle summary from a related YouTube video. With Agent-Reach installed and Twitter cookies ready, the agent calls the corresponding channels to retrieve content. Commands and output are illustrative; actual results depend on platform content.
$ You ›
Summarize the main points of this X thread, then add a subtitle summary
from a related YouTube video.
$ twitter-cli read https://x.com/.../status/...
claude: Agent ›
Read 8 posts in the thread. Main points:
hl: · Claim: Small teams use agents to replace most repetitive back-office work
hl: · Evidence: The author lists 3 automated workflows and the working hours saved
hl: · Dispute: Replies question reliability and review costs
$ yt-dlp --write-auto-sub --skip-download "https://youtu.be/..."
claude: Agent ›
Video subtitle summary (matching the thread's topic):
1. Demonstrates the complete workflow for assigning customer-support classification to an agent
2. Emphasizes that human review remains a required step
3. Provides measurable adoption metrics
$ You › Do the two sources contradict each other?
claude: Agent ›
The thread downplays review, while the video explicitly says human review cannot be skipped.
ok: → Both agree on "saving working hours", but disagree on "how much review can be saved".
The same agent read tweets, extracted subtitles, and compared them to identify disagreements, with zero API fees throughout.
Cross-platform research procedure
The agent can retrieve source material across platforms: it chooses channels according to the task, using twitter-cli for Twitter and yt-dlp for YouTube, then compares the original content directly. You do not need to copy and paste manually or buy an API for each platform.
For research, public-opinion monitoring, and competitor tracking that require substantial reading, requests to the agent can replace manual collection across platforms. The underlying tools remain established open-source projects, with Agent-Reach providing a unified interface.
Usage limits and risks
Multi-channel workflow integration
Once channels work, integrate web reading into your agent workflow. These directions use Agent-Reach's existing capabilities.
Advanced directions
1. Add Agent-Reach to the agent's tool list. In Claude Code, Cursor, Windsurf settings, or CLAUDE.md, specify that web-reading tasks should prefer the relevant Agent-Reach channel so the agent chooses the correct tool automatically.
2. Combine channels for research. Connect search (Exa), social platforms (Twitter/Reddit), video (yt-dlp), and web pages (Jina) in one task so the agent summarizes across sources rather than querying one source at a time.
3. Use RSS for ongoing tracking. Put blogs and news you want to follow into RSS channels, then schedule the agent to read and summarize them regularly for lightweight public-opinion or competitor monitoring.
4. Extend with new channels. The architecture uses pluggable channels (channels/PLATFORM.py). To support a new platform, follow existing channel structures or connect the corresponding MCP server.
5. Share across agents. One installation serves multiple compatible agents (Claude Code, OpenClaw, Cursor, and Windsurf), so switching tools does not require reconfiguring web reading.
Further reading
① GitHub repo: Original source for installation documents, channel lists, and release notes. ② docs/install.md: Installation document to give the agent. ③ agent-reach doctor: Direct diagnostics for checking channel status at any time.
Give your AI agent eyes to see the entire internet. One CLI, zero API fees.