Agent Reach

Open source alternative to Firecrawl, Bright Data and Nango

100.2K stars
MIT
Active this week
Last commit 2 days ago

Last verified Oct 20, 2026

Connect AI agents to web pages, Twitter, Reddit, YouTube, Bilibili, GitHub, and RSS through a local Python CLI with no API fees.

Visit website GitHub100.2K

Read the full breakdownWho it is for, how it works, trade-offs, the quick start and the FAQ

who it's for

Who Agent Reach is for#

Coding agents that need web context

Give Claude Code, Cursor, or Windsurf access to GitHub issues, Twitter discussions, Reddit threads, and web pages as part of a research or debugging task, without wiring up each source separately.

Skip if:

Your agent workflow is fully self-contained in local files with no need for live web or social data.

Social listening and community research

Search Twitter, Reddit, Xiaohongshu, and YouTube for discussions about a product, library, or topic. Useful for developers tracking public sentiment, monitoring bug reports, or gathering feedback before a release.

Skip if:

You need real-time streaming or bulk historical data rather than ad-hoc searches.

RSS and content monitoring

Subscribe to RSS and Atom feeds through feedparser and pull fresh items on demand. Works with any standard feed URL and returns structured data the agent can process directly.

Skip if:

You need push notifications or a persistent background daemon to monitor feeds continuously.

GitHub repository research

Pull public repository details, README content, issue threads, and code search results through the official gh CLI. Useful when an agent needs to understand a dependency, check for known bugs, or compare implementation approaches across repos.

Skip if:

You only need to search code within your own repos, where native IDE integration covers the use case.

the problem

The problem it solves#

AI agents hit access walls the moment a task requires live web or social context. Twitter API access costs money and requires approval. Reddit's anonymous API endpoints have been blocked. YouTube subtitle extraction needs a separate library. Xiaohongshu and Bilibili require active login sessions. The result is that developers either pay for managed scraping APIs or spend hours wiring up credentials for each platform before their agent can do the actual work. Agent Reach addresses this by bundling all those integrations into a single install command, with a routing layer that probes real backends, detects failures, and falls back automatically when a platform changes its access policy.

how Agent Reach solves it

How it solves it#

15 Source Channels

Covers web pages, YouTube, RSS, Exa web search, GitHub, Twitter/X, Bilibili, Reddit, Facebook, Instagram, Xiaohongshu, LinkedIn, V2EX, Xueqiu, and podcast audio. Each channel is a separate backend list with a primary tool and one or more fallbacks.

Zero API Fees

Every upstream tool is open source and free. The only optional cost is a server-side proxy (around $1 per month) when running on a remote machine whose IP gets blocked by social platforms. Local machines do not need a proxy.

Built-in Diagnostics

agent-reach doctor probes each channel in real time, names the active backend, and prints a fix prescription for anything that is not working. There is no guessing about which integrations are ready.

Local Credential Storage

Cookies and tokens are stored in ~/.agent-reach/config.yaml with file permissions set to 600 (owner read-write only). Nothing is uploaded. The project recommends using a dedicated secondary account for platforms that allow scripted access, to limit exposure if a cookie is compromised.

Multi-Backend Routing

Each channel lists a preferred tool and ordered fallbacks. When a platform changes its access policy (for example, when yt-dlp was blocked by Bilibili in June 2026 and the project switched to bili-cli), the routing updates automatically. Users get the change at the next update with no configuration changes required.

strengths · trade-offs

Strengths and trade-offs#

Strengths

Broadest free source coverage15 channels across web, social, video, and code platforms in a single install. No other free local tool covers Twitter, Reddit, YouTube, Bilibili, Xiaohongshu, and GitHub with working backends out of the box.
MIT license with no usage restrictionsThe MIT license lets individuals, teams, and companies use, fork, and modify the code with no attribution or commercial-use restrictions beyond the license text itself.
Actively maintained routing layerThe project tracks upstream tool breakage and updates the channel router when platforms change access policies. The June 2026 Bilibili switch (yt-dlp blocked, bili-cli substituted) showed this process working in practice with no user action required.
Works with any shell-capable agentClaude Code, Cursor, Windsurf, OpenClaw, and any agent that can execute shell commands can call agent-reach commands directly. The install script registers a SKILL.md in the agent's skills directory so the agent knows which commands to call for which task.

Trade-offs

-Social platforms require manual credential setupTwitter, Reddit, Xiaohongshu, Facebook, and Instagram all require cookies or login sessions before the agent can search or read them. The doctor command walks through what is missing, but the credential export step is manual and platform-specific.
-Server deployments need a proxySocial platform APIs block requests from non-residential IPs. Running Agent Reach on a cloud server requires a residential proxy (around $1 per month). Local development machines are not affected.
-Cookie-authenticated accounts carry ban riskPlatforms can detect scripted access from cookie tokens and restrict or ban the account. The project's docs recommend using a dedicated secondary account rather than your main account for any cookie-authenticated channel.

versus alternatives

Agent Reach vs alternatives#

Agent Reach is an MIT-licensed, local-first alternative to Firecrawl, Bright Data, Nango, and Apify for giving AI agents web and social context. It is best for individual developers and small teams that want a CLI an agent can call to read web pages, search social sources, pull YouTube or RSS context, and inspect GitHub without paying for platform access.

vs Firecrawl

Firecrawl is a hosted API for web scraping, crawling, search, site mapping, monitoring, and browser interaction, with usage priced around credits per page, search result, or browser minute. Agent Reach covers a wider range of personal research channels (Twitter, Reddit, YouTube, Bilibili, Xiaohongshu) that Firecrawl does not target, but it is not a production scraping API with SLAs, managed rate limits, or support. Agent Reach is the better fit when an agent needs to gather context from social and video platforms. Firecrawl is the better fit when your application needs reliable, scalable web extraction infrastructure.

vs Bright Data

Bright Data is an enterprise-grade data platform with managed proxies, browser automation, pre-built scrapers, and dataset products, priced at commercial rates for scale and uptime. Agent Reach is lighter and more local: it chains open-source tools and local cookies so an agent can gather context without purchasing proxy capacity or commercial data products. Bright Data wins when you need managed scale, professional support, or commercial compliance. Agent Reach wins when cost is the constraint and the use case is personal research rather than production data pipelines.

vs Nango

Nango centralizes API authentication and data syncing for product integrations: OAuth flows, token refresh, and pre-built API connectors for hundreds of services. It solves the infrastructure problem of connecting your product to third-party APIs at scale. Agent Reach is a different kind of tool: it handles agent-side read access to public and cookie-authenticated sources, not the OAuth and sync infrastructure that product developers integrate into their backend. The two tools rarely compete directly. Nango is for product builders who need managed API auth. Agent Reach is for agent builders who need local web context.

vs Apify

Apify is a hosted platform for web scraping and browser automation built around Actors: containerized scraping and automation scripts you run or rent from a marketplace of over 66,000 ready-made tasks. It includes managed residential and datacenter proxies, a full cloud runtime, and paid plans starting at $29 a month with usage billed by compute unit (1 GB of RAM per hour). Agent Reach takes a different approach: no cloud runtime, no marketplace, no proxy fees. You run it locally, it chains open-source CLI tools, and your credentials stay in a local config file. Apify is the better fit when you need a production-grade, managed scraping pipeline with proxy rotation and a large library of prebuilt scrapers. Agent Reach is the better fit when you want a free, local tool for feeding an AI agent context from public web, social, and video sources without standing up infrastructure or paying per run.

install · quick start

Quick start#

bash

Agent Reach installs as a Python package; run the doctor command after install to check and configure your channels.
```bash
pip install agent-reach
agent-reach doctor
```

tech stack · detected from GitHub

What it's built on#

Languages: Python

frequently asked

FAQ#

How does Agent Reach access Twitter without paying for the API?

Agent Reach uses twitter-cli with cookie-based authentication. You export your session cookies from a logged-in browser using a browser extension, then configure Agent Reach with those cookies. The tool uses your existing session to read and search Twitter rather than calling the paid API. The project recommends using a secondary account because scripted access carries a platform detection and ban risk.

What does Agent Reach actually cost?

Nothing for local use. All upstream tools are open source and free. The one optional cost is a residential proxy (around $1 per month) if you run Agent Reach on a cloud server whose IP gets blocked by social platforms. That cost is optional and does not apply to local machines.

Which sources work without any configuration after install?

Web pages (via Jina Reader), YouTube transcripts and search (via yt-dlp), RSS and Atom feeds (via feedparser), public GitHub repositories (via gh CLI), and Bilibili search and video details (via bili-cli) all work immediately after install with no credentials required. Full-web semantic search through Exa also configures automatically during install.

Does it work on a remote server, not just a local machine?

Yes, with one extra step: social platforms block server IPs, so you need a residential proxy (around $1 per month) for cookie-authenticated channels like Twitter and Reddit. The agent-reach doctor command detects whether you are on a server and shows the proxy configuration steps. Channels that do not require authentication (web, YouTube, RSS, GitHub, Bilibili) work on a server without a proxy.

What happens when a platform changes its access method?

Agent Reach maintains a routing file for each channel with an ordered list of backends. When a tool stops working (the project gives Bilibili as an example: yt-dlp was blocked by Bilibili's anti-scraping system in June 2026 and replaced with bili-cli), the project updates the routing and you get the fix at the next update with no configuration changes on your end.

also worth a look