Index Everything, or Read Everything? The Dilemma of Feeding Specs to AI in Multi-Repo Development

Short summary

A developer explores the dilemma of feeding specification documents from separate repositories to AI coding assistants like Cursor. Multi-root workspaces pull in irrelevant files and inflate token usage, while manual file fetching wastes tokens on search. The author ultimately discovers GitHub MCP Server's remote-hosted GA version as a solution, avoiding local Docker setup. The article captures a real workflow problem for microservice developers using AI-assisted coding.

• •Multi-root workspaces in Cursor add noise and token cost when you only need a few spec files
• •Manual file fetching wastes tokens loading full file contents just to search them
• •GitHub MCP Server's remote-hosted GA version solves cross-repo spec access without local infrastructure

Generated with AI, which can make mistakes.