Agent-Reach Lets AI Agents Read Twitter and Reddit Without API Fees

What is Agent-Reach?

Agent-Reach is a free, open-source Python program that lets AI agents read sites that usually shut programs out, such as X, Reddit and YouTube. AI agents are chatbots that can run commands on your computer, like Claude Code or Cursor.

Agent-Reach's GitHub description promises "one CLI, zero API fees." A CLI is a program you run by typing in a terminal. An API fee is what a site charges programs for official access.

Agent-Reach spread fast. The repo had 90,114 stars on GitHub on 2026-10-04, about seven months after it was created in February, and it picked up 3,952 of them in its week on GitHub's trending page. A developer who goes by Panniantong wrote it. It's MIT-licensed, and the last update was 2026-09-15.

The Agent-Reach README's pitch is about money and walls. It says moderate use of X's official API costs about $215 a month, and that Reddit simply refuses requests coming from server addresses.

How does Agent-Reach work?

Agent-Reach doesn't do the reading itself. It picks a free tool for each site, installs it and keeps checking that each one still works, then tells your AI agent which tool to call for which site. The README calls this "a capability layer." Those instructions live in a cheat sheet, a file called SKILL.md.

Here's what that looks like. You paste a YouTube link into Claude Code and ask what the video covers. The agent checks its cheat sheet, runs yt-dlp --dump-json on the link to pull the subtitles, and sums them up. You never type a command. For a web page it calls Jina Reader, which turns any URL into clean text.

In Agent-Reach, each site gets a ranked list of backup tools. When the top one breaks, the next moves up. The README's own example: in June 2026 Bilibili, a Chinese video site, started blocking yt-dlp, so Agent-Reach switched to a tool called bili-cli. Running agent-reach doctor shows which tool each site is using and what's broken.

Not every site in Agent-Reach is free in the same way. The README lists YouTube, GitHub, RSS feeds and web pages as "zero config," and Bilibili gets the same label. X search and Reddit are different. Both need a cookie from your own account, the small login token your browser keeps after you sign in. Facebook and Instagram only work on a desktop computer, where a tool called OpenCLI borrows your logged-in Chrome session.

Agent-Reach's README contradicts itself on web search. Its platform table says Exa search needs no API key, but its sample doctor output tells you to sign up at exa.ai for a free one.

How do you use Agent-Reach?

The main way to install Agent-Reach is a sentence, not a command. You paste it into your AI agent, which fetches the instructions at the link and follows them. A manual install with pipx and a Claude Code Skill also work. Here's the sentence:

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

So your agent runs steps from a file someone else wrote. Read install.md before you paste it. The README says the default install only checks your machine, and system changes need an explicit --system flag. Approve that flag yourself.

To install Agent-Reach by hand on a Mac or Linux machine, install.md recommends pipx, a tool that installs Python programs in their own space. The second line runs the read-only check:

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

If Python stops with an "externally-managed-environment" error, which install.md says happens with Homebrew's Python, use a virtual environment (a private Python folder) instead:

python3 -m venv ~/.agent-reach-venv
source ~/.agent-reach-venv/bin/activate
pip install https://github.com/Panniantong/agent-reach/archive/main.zip
agent-reach install --env=auto

Once you've seen what Agent-Reach wants to change, apply the system changes. Then check what's working:

agent-reach install --env=auto --system
agent-reach doctor

Claude Code users can install Agent-Reach as a Skill instead. The README says the agent will then install the CLI itself if it's missing:

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

X search in Agent-Reach is opt-in. The project's Twitter guide gives these steps:

1. 1Install twitter-cli with pipx install twitter-cli.
2. 2Install the Cookie-Editor browser extension and log in to x.com, ideally with a spare account.
3. 3In Cookie-Editor, choose Export, then Header String.
4. 4Run agent-reach configure twitter-cookies and paste it in. It saves two values, auth_token and ct0, to ~/.agent-reach/config.yaml.

On a Linux server, skip Agent-Reach's OpenCLI routes, since they need a desktop browser. Reddit there means rdt-cli, after you run rdt login.

Agent-Reach vs paid options: what does it cost?

Agent-Reach itself is free. Its README says the only possible cost is a $1-a-month proxy, needed only on networks that block Reddit or X. Paid alternatives run from X's official API at $0.005 per post read up to Bright Data's $499-a-month Scale plan.

News Desk checked these vendor pages on 2026-10-04. The X figure of $215 a month comes from Agent-Reach's README, not from X.

The YouTube gap is why yt-dlp is in Agent-Reach's kit. Google's free YouTube Data API won't give you captions for a video unless you have permission to edit it. And none of the cheap search APIs reach inside logged-in X or Reddit.

What are Agent-Reach's limits?

Agent-Reach's main limits are setup, breakage and hardware. X search and Reddit need your own login, the free tools snap when sites change, and Facebook and Instagram only work on a desktop computer.

For X search and Reddit, every request goes out as you. That means exporting a cookie from your browser for X, and running rdt login for Reddit on a server.

Then there's breakage. The tools Agent-Reach installs copy how a site behaves on the inside instead of using a stable official API, so they snap when the site changes. The README gives examples: Bilibili blocked yt-dlp in June 2026, and a batch of single-site tools were abandoned in March. Backup lists soften that. They don't stop it.

Some Agent-Reach routes need a desktop. Facebook and Instagram go through OpenCLI, which borrows your logged-in Chrome session. On a Linux server, they're out.

Agent-Reach's README also lists sponsors. BrowserAct and Tencent Cloud are on the list. So are CoreClaw and AstraFlow. All four sell paid scraping or cloud products.

Who is Agent-Reach for?

Agent-Reach covers the zero-config uses: reading YouTube subtitles, GitHub repos, RSS feeds and web pages. Those cost nothing, and none of them logs in as you. With Claude Code on a Mac it's a short install, and agent-reach doctor shows what's working.

The official X API covers X data without the breakage, since it doesn't snap when the site changes. At $0.005 a read, 10,000 posts cost about $50. For Reddit, both paid routes come with conditions. Reddit's own API needs manual approval, and Apify's Reddit scrapers ran about $0.60 to $4 per 1,000 results, with at least one already marked deprecated.