CodeSpine
MCP server for AI agents to understand Java codebase
Start new thread
Vinayak S

CodeSpine - MCP server for AI agents to understand Java codebase

by•
CodeSpine is a Python MCP server that indexes your Java project once and serves structural info to AI agents on demand. Instead of reading 30 files per query, it makes one MCP call and returns caller trees, impact paths, dead code reports, and execution flows. Features: hybrid BM25+fuzzy search, git change coupling, multi-module support, dirty overlay for in-progress edits. Zero hallucinations, 90% fewer tokens. MIT licensed.

Add a comment

Replies

Best
Vinayak S
Maker
📌
Hey Product Hunt! I built CodeSpine after getting frustrated with AI coding assistants hallucinating method signatures and missing cross-module dependencies in my Java monorepo. The core insight: AI agents don't need to read your entire codebase — they need a structured index they can query on demand. CodeSpine indexes your project once and exposes it via MCP (Model Context Protocol), so instead of feeding 30 files into context per query, the agent makes one call and gets exactly what it needs: caller trees, impact paths, dead code reports, and execution flows. Results on my own projects: ~90% fewer tokens per query and zero hallucinated method signatures. MIT licensed and works with any MCP-compatible AI agent (Claude, Cursor, etc.). Would love feedback from anyone working with large Java codebases! GitHub: https://github.com/vinayak3022/c... LinkedIn: https://www.linkedin.com/in/thev...