📑Table of Contents

  1. Why it exists

  2. What it can actually do

  3. Before you start

  4. Installation

  5. Setup step by step

  6. Prompts to use it

  7. Commands table

  8. Daily workflow

  9. Tips

  10. Troubleshooting

  11. Quick reference

🧩 Why it exists

You spin up a new AI agent and want it to read the actual internet. Not a sanitized search snippet, the real conversation: the Twitter thread where people are arguing, the Reddit comments where the bug actually got solved, the YouTube tutorial with the steps in the captions.

The problem is every one of those platforms has a wall. Twitter wants an API subscription that runs around $215 a month for moderate use. Reddit started returning 403 to server and datacenter IPs. Bilibili blocks overseas IPs. XiaoHongShu wants you logged in. So every time you set up a new agent, you go hunting for tools, install a pile of dependencies, debug configs, and then a few months later one of those single-platform CLIs goes unmaintained and you do it all again.

Agent Reach collapses that into one step. It picks the most reliable access path for each platform, installs it, and runs a health check. After that, your agent talks to the underlying tools directly. There is no wrapper sitting in the middle. When an access path stops working, the project re-routes to a new one and you don't rewrite your code.

⚙️ What it can actually do

You say

What happens

"Read this link"

Fetches the page as clean Markdown via Jina Reader

"What's this GitHub repo about?"

Runs gh repo view owner/repo

"Search GitHub for LLM frameworks"

Runs gh search repos "LLM framework"

"Read this tweet"

Runs twitter tweet URL (cookie unlocks search + timeline)

"Search Reddit for X and read the top thread"

Runs rdt search then rdt read POST_ID

"What does this video cover?"

Pulls subtitles + metadata with yt-dlp

"Subscribe to this RSS feed"

Parses it with feedparser

"Search XiaoHongShu for X"

Runs xhs search after cookie login

❝

⚠️ Capabilities per platform differ. Some are read-only, some add search, a few (XiaoHongShu) add post/comment/like. Check agent-reach doctor for what's live on your machine.

✅ Before you start

  • Python 3.10 or newer installed

  • An AI agent that can run shell commands (Claude Code, Cursor, Windsurf, OpenClaw, Codex, etc.)

  • A browser logged into any platform you want cookie access for (Twitter, Reddit, XiaoHongShu)

  • Optional: a free Exa API key for web semantic search (sign up at exa.ai)

  • Optional: a residential proxy (~$1/month, e.g. Webshare) only if you need Bilibili from a server. Local machines don't need this.

❝

⚠️ This is built for agentic research and prototyping. It is not meant for production web scraping at scale.

💻 Installation

The intended path is to hand the install prompt to your agent and let it do the work. There are also manual options.

Easiest: let your agent install it

Paste this to your agent (Claude Code, Cursor, OpenClaw, etc.):

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

Security-conscious? Use safe mode, which won't auto-install system packages:

Install Agent Reach (safe mode): https://raw.githubusercontent.com/Panniantong/agent-reach/main/docs/install.md  install with the --safe flag

Manual install (Mac / Linux / Windows)

The documented manual command pulls the archive directly, so it's the same across operating systems as long as Python 3.10+ and pip are present:

bash

pip install https://github.com/Panniantong/agent-reach/archive/main.zip
agent-reach install --env=auto

Mac notes: if pip points at system Python, use pip3. Homebrew Python recommended.

Windows notes: run from PowerShell or a terminal where Python is on PATH. Use pip or py -m pip.

Linux notes: you may need pip3. On a server, remember Bilibili and some platforms need a proxy.

Install as a Skill (Claude Code / OpenClaw / agents with Skills support)

bash

npx skills add Panniantong/Agent-Reach@agent-reach
❝

⚠️ The plain pip install agent-reach shorthand appears in the repo's FAQ, but the primary documented manual command is the archive URL above. If the short form fails, use the archive URL.

Update later

Update Agent Reach: https://raw.githubusercontent.com/Panniantong/agent-reach/main/docs/update.md

Or manually:

bash

pip install --upgrade https://github.com/Panniantong/agent-reach/archive/main.zip

🪜 Setup step by step

  1. Confirm Python: run python --version (or python3 --version) and check it's 3.10+.

  2. Install Agent Reach using one of the methods above.

  3. Run the installer: agent-reach install --env=auto. It detects your environment and sets up upstream tools.

  4. Run the health check: agent-reach doctor. It prints every channel and whether it's ready, configurable, or blocked.

  5. Read the output. Anything marked ✅ works right now with zero config (GitHub, YouTube, RSS, web pages).

  6. For anything marked ⬜ or "Cookie," configure it on demand (next section). Skip platforms you don't need.

  7. Tell your agent to read its SKILL.md so it knows which commands map to which request. You don't memorize commands; the agent does.

🧪 Prompts to use it

Reading and research

Read this link and summarize it: <URL>
What does this YouTube video cover? Pull the subtitles: <URL>
What is this GitHub repo about and how many stars does it have? <repo URL>

Search across platforms

Search GitHub for open source RAG frameworks and list the top 5 by stars.
Search Reddit for recent threads about <topic> and summarize the top comments.
Search Twitter for what people are saying about <topic> this week.

Competitive / content radar

Pull the latest tweets from <handle>, the top Reddit thread on <topic>,
and the top GitHub repo for <category>. Summarize what's trending.

📟 Commands table

Command

What it does

agent-reach install --env=auto

Installs upstream tools and detects environment

agent-reach doctor

Shows each channel's status and how to fix issues

agent-reach skill --install

Registers the SKILL.md so your agent auto-knows commands

gh repo view owner/repo

Reads a GitHub repo

gh search repos "query"

Searches GitHub repos

twitter tweet URL

Reads a tweet

twitter search "query" -n 10

Searches Twitter (needs cookie)

rdt login

Logs into Reddit via browser cookies

rdt search "query"

Searches Reddit

rdt read POST_ID

Reads a full Reddit post + comments

yt-dlp --write-sub --skip-download "URL"

Pulls YouTube/Bilibili subtitles

xhs login

Logs into XiaoHongShu

xhs search "query"

Searches XiaoHongShu notes

curl https://r.jina.ai/URL

Reads any web page as Markdown

📆 Daily workflow

A content/research operator using this day to day usually looks like:

Morning, you ask your agent for a trend sweep: latest tweets from a few handles, the top Reddit thread in a subreddit, the top new GitHub repos in a category. The agent runs twitter search, rdt search, and gh search repos and hands back a single summary. No tabs, no API dashboards.

When something looks worth a deeper read, you point the agent at the specific link. It pulls the full tweet thread, the Reddit comments, or the YouTube captions, and you get the substance without leaving your chat.

Once a week you run agent-reach doctor to make sure nothing broke after an upstream tool update. If a channel went red, the doctor output tells you exactly what to fix.

💡 Tips

  • Run agent-reach doctor first, every time something feels off. It tells you which backend each platform is currently routed to, so you're never guessing.

  • Only configure the platforms you actually use. Every cookie/login step is optional. A minimal setup with just the zero-config channels (GitHub, YouTube, RSS, web) is genuinely useful on its own.

  • Cookies stay local in ~/.agent-reach. They're never uploaded. If you're privacy-conscious, use safe mode during install so it won't touch system packages.

  • Local machine vs server matters. Reddit and Bilibili behave differently from a datacenter IP. If you're on a server and a channel 403's, that's usually why, and a cheap proxy fixes it.

  • Because the project re-routes when an upstream tool dies, pin nothing manually. Let doctor tell you what's current rather than hardcoding a backend.

🛠️ Troubleshooting

❝

⚠️ Reddit returns 403 from a server / datacenter IP. Reddit requires auth for all API requests since 2024. Fix: run rdt login (it auto-extracts cookies from your browser). On a server you may also need a residential proxy.

❝

⚠️ Bilibili video info fails on a server. Server IPs are blocked. Fix: configure a proxy (~$1/month). Local machines work directly with no proxy.

❝

⚠️ Twitter search returns nothing. Search needs a cookie. Fix: make sure you're logged into x.com in your browser, then tell your agent "help me configure Twitter cookies," or run the cookie export step.

❝

⚠️ pip install agent-reach fails or installs the wrong thing. Use the documented archive URL instead: pip install https://github.com/Panniantong/agent-reach/archive/main.zip.

❝

⚠️ Web semantic search shows ⬜ in doctor. You need a free Exa key. Fix: sign up at exa.ai, then add the key when your agent prompts for it.

🔖 Quick reference

bash

# install (manual)
pip install https://github.com/Panniantong/agent-reach/archive/main.zip
agent-reach install --env=auto

# health check (run this whenever something breaks)
agent-reach doctor

# core reads
gh repo view owner/repo
gh search repos "query"
twitter tweet URL
twitter search "query" -n 10
rdt login && rdt search "query"
rdt read POST_ID
yt-dlp --write-sub --skip-download "URL"
xhs login && xhs search "query"
curl https://r.jina.ai/URL

# update
pip install --upgrade https://github.com/Panniantong/agent-reach/archive/main.zip

📋 Copy-paste install prompt (hand this to your agent)

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

After install, run `agent-reach doctor` and report which channels are ready,
which are configurable, and which are blocked. Then read SKILL.md so you know
which command maps to each platform. Don't configure platforms I haven't asked for.

By The AI Leverage - Learn and master AI daily

Keep Reading