When building AI agents, I struggled with web access tools: most returned raw HTML, cluttered markdown full of navs/footers, or failed on JavaScript-heavy pages.
I reverse-engineered how Claude Code's tools handle it (
https://medium.com/@liranyoffe/reverse-engineering-claude-co...),
then built Quercle.dev: a simple, vendor-agnostic API that renders JS, extracts ultra-clean LLM-ready content, synthesizes answers with citations, and minimizes tokens.
Key features:
- Fetch: URL + prompt → concise, structured result (optimized for context windows)
- Search: Query → synthesized answer w/ verifiable sources
- Python/TS SDKs, LangChain/AI-SDK and MCP integrations
Sign up for 50 free credits to try it out!
Would love feedback from the HN community!