After spending years building code analysis and migration tools at Mobilize.Net(now part of Snowflake), and now working with AI developer tools, I've learned that the real magic happens when you can bridge human expertise with AI capabilities. That's exactly what led me to create roberto-mcp - a Model Context Protocol server that transforms how AI understands codebases(name picked by my son honoring our dogs name).
The Journey: From Enterprise Tools to AI Integration
My background spans building migration tools, syntax parsers, and performance optimization systems. At Mobilize.Net, I learned the intricacies of code parsing and automated manipulation across multiple languages.
Then most recently working at AI Developer tools at Amazon, I quicky find out that there were constraints and usually back and forth with AI tools if they didn't have the right context to execute a task.
But here's what I realized: all this expertise in understanding code structure was sitting in isolated tools. Meanwhile, AI assistants were struggling with code comprehension tasks when analyzing code bases at large scale.
Enter the Model Context Protocol
When I discovered Amazon Q CLI and its MCP integration capabilities, something clicked. Here was a standardized way to give AI tools deep, contextual understanding of codebases. Instead of AI trying to parse code from scratch every time, I could provide it with the structured, semantic understanding I'd been building for years.
Building Roberto-MCP: Performance as a top priority
I chose Rust for one simple reason: when you're indexing massive codebases, performance isn't optional. The server delivers:
• Lightning-fast symbol lookups - Because waiting kills developer flow
• High-speed indexing - Handle enterprise-scale repositories efficiently
• Intelligent caching - Quick restarts for previously analyzed projects
The Real Impact: Natural Language Code Queries
Instead of grepping through thousands of files, developers can now ask:
• "Find all error handling patterns in this codebase"
• "Show me functions that implement binary search algorithms"
• "Get an overview of all API endpoints in this service"
The AI gets structured, contextualized data instead of raw text dumps.
Technical Architecture That Scales
Built on Tree-sitter for consistent parsing across 15+ languages, with:
• Binary persistence for instant restarts
• Automatic LRU eviction for memory management
• SHA-256 change detection for incremental updates
• BM25 statistical search through code content
Open source
You can checkout RobertoMCP code at https://github.com/kensave/roberto-mcp.
Conclusion
The future of developer tools isn't just about better AI - it's about giving AI the right context to understand what we're actually trying to build.
Results
Dive deep unknown code bases, in seconds.
Prompt sample:
LLMs are very powerful, but this aims to give an small push to it in the right direction when needed.
Top comments (2)
Oh man. I have developed something much more primitive, but absolutely aligned to this. I was looking for how to lift it to the next step, this might be exactly it.
Contextslicer.web.app
Context management is a very open ended and interestic topic right now :)