VIBETTER
Open-Source MCP Server for AI Coding IDEs
Open-source MCP server powering codebase-intelligent AI workflows in Claude Code, Cursor, and Windsurf.
Context
VIBETTER is an open-source MCP (Model Context Protocol) server designed to give AI coding assistants genuine codebase understanding. Rather than treating code as flat text, VIBETTER provides structured, source-grounded intelligence that AI tools can use to explain changes, answer questions with file:line precision, debug contextually, and visualize dependencies.
The problem
AI coding assistants like Claude Code, Cursor, and Windsurf operate on code context, but they lack structured understanding of codebases — how files relate, what changed and why, and where specific logic lives. Without grounding, AI responses about code are often generic or hallucinated. The challenge was building an MCP server that provides this structured intelligence reliably, with fast response times and graceful degradation.
How I built it
Designed the server architecture around FastMCP 3.0 for clean tool registration and protocol compliance
Integrated Gemini API with long-context support for analyzing large diffs and multi-file relationships
Implemented fallback model chaining to ensure responses even when primary models are unavailable
Built Watchdog-based file monitoring for automatic context refresh as the codebase evolves
Structured all outputs with file:line citations for source-grounded, verifiable responses
Why these choices
FastMCP 3.0 over custom protocol
Standardized MCP protocol ensures compatibility across IDEs while reducing maintenance burden. FastMCP's tool registration pattern keeps the codebase clean and extensible.
Gemini API with long context
Large diffs and multi-file analysis require substantial context windows. Gemini's long-context capability handles this natively without complex chunking.
Fallback model chaining
Production reliability requires graceful degradation. If the primary model fails or hits rate limits, the server automatically falls back to alternative models.
Source-grounded citations
Every response references specific files and line numbers, making outputs verifiable and reducing the risk of hallucinated code explanations.