Post by Muhammad Arbab Arshad

The current OpenAI complete documentation stands at about 336,000 tokens. For any question, AI coding assistant needs maybe 300 of them. MCP services like Context7 do a good job of retrieving the right chunks, but you're still looking at 5,000 to 10,000 tokens each time. That adds up fast and causes a lot of context pollution. But thinking about it from a Pareto principle point of view, 90% of the use cases could be solved by about 10% of the documentation. That's the inspiration I drew to build Context45. I curated the Anthropic and OpenAI docs down to just the core patterns people actually use day to day, then added semantic search on top. Now you get 100 to 300 tokens per query instead of thousands. I'm releasing the beta version and invite the community to build on top of it. Add more libraries, improve the docs, and have fun with it. I got the chance to build the MCP server, npm package, the landing page, and the whole GitHub repository within a day with the latest Claude Code Opus 4.6. It's been a fun experience. https://context45.com/