I got frustrated watching Claude Code fail at using modern APIs (ask it about GPT-5 and it says it doesn't exist). Existing solutions like Context7 dump thousands of tokens of irrelevant docs into context. So I built DeepCon.

How it works:

- Crawled 10,000+ official docs using agentic browser automation and structured them hierarchically

- Query decomposer breaks down requests, searches in parallel, then merges only relevant context

- Returns just what's needed: 2.4x fewer tokens than Context7

Results on our benchmark:

DeepCon achieved 90% accuracy vs Context7's 65% on real-world tasks with modern AI frameworks. Without any MCP context, Sonnet 4.5 scored 0%.

It's an MCP tool. You can just plug it into Claude Code/Cursor and suddenly they understand the latest libraries and APIs.

GitHub benchmark: https://github.com/opactorai/context-bench

Service: https://deepcon.ai

Built this because I needed it. Would love feedback!