Agent Reach is a capability layer that equips AI agents with ready‑to‑use internet access by automatically selecting, installing, and maintaining the most reliable backend for each supported platform. Zero‑configuration channels include web page reading via Jina Reader, YouTube subtitle extraction and search with yt‑dlp (154 K ★), RSS parsing through feedparser, and generic web search using Exa’s semantic search accessed through mcporter, which requires no API key. For platforms that need authentication—Twitter/X, Reddit, Facebook, Instagram, LinkedIn, B站, 小红书, 雪球, 小宇宙播客, Boss直聘—Agent Reach guides the agent to configure login state via OpenCLI, twitter‑cli, rdt‑cli, xiaohongshu‑mcp, bili‑cli, gh CLI, or mcp‑server‑linkedin, storing cookies and tokens locally in ~/.agent‑reach/config.yaml with file permissions 600. The toolset includes a doctor command (agent‑reach doctor) that reports channel status and the active backend, and an install command that can run in dry‑run mode, optionally install system dependencies with --system, and register skills in the agent’s SKILL.md file. Updates are performed with a single URL, and the project is fully open source, free of charge aside from an optional ~$1/month proxy for server deployments. Compatible agents include Claude Code, OpenClaw, Cursor, Windsurf, and any agent capable of executing shell commands.
Read original
github-trending/python