Agentfetch

Wraps multiple web fetchers (Trafilatura, Jina Reader, FireCrawl, pypdf) behind a single MCP interface that routes requests based on URL type and keeps you under token budgets. Exposes four tools: fetch_url for single pages with token caps and caching, estimate_tokens to check size before committing context, fetch_multiple for up to 20 concurrent requests, and search_and_fetch for web search plus automatic result retrieval. Handles JS-heavy sites through FireCrawl, extracts PDFs locally, and falls back through a cost hierarchy when one fetcher fails. The routing is automatic, so you call one tool instead of wiring up four separate scrapers and a Redis layer yourself. Useful when your agent needs web content but you want to avoid blowing context windows on 50,000-token blog posts.