Scout is a free, self-hosted tool that connects any AI assistant to 9 real-time intelligence sources simultaneously: Reddit, Hacker News, GitHub, YouTube, X, Instagram, Polymarket, 16+ RSS feeds, and general web search. Results from all sources are ranked together by cross-source consensus and returned directly inside your AI conversation as structured tool calls. One-command install. Works with Claude, Cursor, Windsurf, and any MCP-compatible AI client.

When you ask an AI assistant to research a topic, you get a generic web search result.

What you don't get: the Hacker News thread developers are actively debating, the GitHub repository that 800 people starred this week, the Reddit discussion where practitioners are sharing real-world experience, and the Polymarket prediction on the same topic - all ranked together in one response.

Existing tools either search the open web (shallow, generic) or require you to manually check each community one by one. Neither approach tells you where the real signal is.

A 12-tool MCP server that gives any AI assistant access to 9 intelligence sources in a single tool call. When your AI needs to research a topic, it calls Scout directly - no tab switching, no copy-pasting, no manual research loop.

Sources covered: Reddit (10 curated subreddits), Hacker News, GitHub repositories and trending, YouTube channels and search, X/Twitter, Instagram hashtags, Polymarket prediction markets, 16 RSS feeds across tech news and AI research, and general web search via SearXNG or Brave Search API.

The intelligence layer is what separates Scout from a basic scraper. Every result from every source passes through engagement scoring, relevance ranking, near-duplicate removal, and cross-source fusion before being returned. A topic appearing on both Hacker News and GitHub trending ranks higher than the same topic on either source alone. Consensus signal surfaces automatically.

All results are stored locally in a SQLite database with full-text search, so your AI can search history instantly without hitting the network again. Nothing leaves your machine.

There is also a video intelligence pipeline: any YouTube or Instagram video URL can be passed to Scout, which downloads it and transcribes it locally. Returns a full transcript with timestamped segments. No external transcription API. No cost per transcription.

Add this block to your AI client config (Claude Desktop, Cursor, Windsurf, or any MCP-compatible client) and restart:

For Claude Code: claude mcp add scout -- npx pathrix-scout

Reddit, Hacker News, Polymarket, and all 16 RSS feeds need no setup at all. GitHub requires one free login: brew install gh && gh auth login. Everything else is optional.

Cross-source consensus is more valuable than depth on any single source. A topic appearing on Hacker News, GitHub trending, and two subreddits simultaneously is a strong signal. The same topic on one source is noise. Building the cross-source fusion layer before adding more sources was the right call.

Separating the scraping layer from the intelligence layer completely is what makes the system maintainable. Sources break, get rate-limited, or change their structure. None of that should touch the ranking logic.

Zero-dependency core setup is critical for open source adoption. Every required API key or install step is a drop-off point. Starting with public APIs only means anyone can run it in under two minutes without signing up for anything.

MCP is the right protocol for this kind of tool. The alternatives - a browser extension, a sidebar, a separate app - all break the AI workflow. Tool calls inside the conversation are the only approach that does not interrupt how you work.

TypeScript, Node.js, MCP SDK, SQLite (WAL mode + FTS5), Python (instagrapi for Instagram), yt-dlp, faster-whisper, SearXNG / Brave Search API

GitHub: https://github.com/gauravsaxena1997/pathrix-scout-mcp Install: npx pathrix-scout