Upstash Context7 Ships Architecture Overhaul — 65% Token Reduction for MCP Documentation Platform

GitHub·medium signal

Context7, the MCP server that injects up-to-date code documentation into LLM context, shipped an architecture overhaul achieving 65% reduction in tokens (from ~9.7K to ~3.3K average), 38% latency improvement (24s to 15s), and 30% fewer tool calls. The platform serves real-time documentation for any library directly into AI code editors like Cursor and Windsurf via the Model Context Protocol. At 53.8K stars, Context7 has become one of the most-installed MCP servers in the ecosystem.