Announcing the LeanMCP Track Winners at NexHacks Carnegie Mellon!

Answer First

Who won the LeanMCP track at NexHacks Carnegie Mellon? Three outstanding projects: WashedMCP (1st - semantic code search with 70% token savings), neoCortex (2nd - GitHub repository analysis with 85% token savings), and Checkr (3rd - AI mental health companion with cross-modal support).

Key innovations:

• Token optimization achieving 70-85% savings in code intelligence tools
• Cross-modal agentic workflows for healthcare applications
• Advanced semantic search with dependency graph analysis
• LeanMCP integration enabling sophisticated tool calling patterns

Don't miss this: These projects showcase how LeanMCP enables both developer productivity tools and sensitive domain applications like healthcare.

Definition Box

NexHacks: Hackathon hosted at Carnegie Mellon University featuring multiple technology tracks including LeanMCP.

Token Optimization: Techniques to reduce the number of tokens required for AI model interactions while maintaining functionality.

Semantic Code Search: Advanced code search that understands meaning and relationships rather than just text matching.

Cross-modal AI: AI systems that can process and respond through multiple communication channels (text, voice, etc.).

Dependency Graph: Visual representation of relationships between code components showing callers and callees.

Agentic Workflows: AI systems that can take autonomous actions and make decisions within defined parameters.

NexHacks LeanMCP Track Winners

We're thrilled to announce the top three winners from the LeanMCP track at NexHacks hosted at Carnegie Mellon University!

These outstanding projects demonstrated creative and practical applications of LeanMCP - pushing the boundaries of context-efficient tool calling, semantic code understanding, and agentic AI systems.

1st Place: WashedMCP

Built by Pavan Kumar and his team

WashedMCP delivers a token-efficient semantic code search server that returns not just a matching function or class, but also its callers, callees, and relevant same-file context in a single query. By using Tree-sitter parsing, ChromaDB vector storage, and a custom TOON (Token-Optimized Object Notation), it dramatically reduces token usage (up to ~70% savings) while preserving critical relationships. It also features a LeanMCP-powered hook system that learns from repeated tool calls and auto-recommends/installs appropriate MCP tools.

Key Features:

• Token-efficient semantic code search with up to 70% savings
• Tree-sitter parsing for accurate code analysis
• ChromaDB vector storage for fast retrieval
• Custom TOON (Token-Optimized Object Notation) format
• Auto-learning hook system for MCP tool recommendations

2nd Place: neoCortex

Built by Rusheel Sharma and his team

neoCortex is a powerful semantic code retrieval engine for GitHub repositories. It builds a bidirectional dependency graph using Tree-sitter, combines OpenAI embeddings with keyword and graph traversal for hybrid search, and intelligently compresses results to fit within token budgets (achieving 85%+ token savings on average). Fully exposed as a LeanMCP server, it includes tools like search_code, resolve_symbol, and classify_query - making it a seamless companion for any MCP-enabled AI coding assistant.

Key Features:

• Bidirectional dependency graph construction
• Hybrid search combining embeddings, keywords, and graph traversal
• 85%+ token savings through intelligent compression
• Complete LeanMCP server implementation
• Tools: search_code, resolve_symbol, classify_query

GitHub: https://github.com/rusheelsharma/neocortex

3rd Place: Checkr

Built by Jacob Mazelin and his team

Checkr is an empathetic, agentic AI mental health companion built with LiveKit agentic workflows and LeanMCP for shared context across text and voice. It handles proactive onboarding, real-time therapist/resource discovery, automated scheduling with calendar integration, and seamless handoff to human therapists - all while maintaining emotional intelligence and user context across modalities.

Key Features:

• Cross-modal AI companion (text and voice)
• LiveKit agentic workflows integration
• Real-time therapist and resource discovery
• Automated calendar scheduling
• Seamless handoff to human therapists
• Emotional intelligence with persistent context

Final Thoughts

These winners beautifully illustrate the versatility of LeanMCP: from dramatically improving token efficiency and context quality in code intelligence tools, to enabling sophisticated, cross-modal agentic applications in mental health support.

The projects showcase three distinct use cases:

1. Developer Tools: WashedMCP and neoCortex both tackle the challenge of making AI coding assistants more efficient and context-aware
2. Healthcare AI: Checkr demonstrates how LeanMCP enables complex, multi-modal agentic workflows in sensitive domains
3. Token Optimization: All three projects achieved significant token savings while maintaining or improving functionality

A huge congratulations to these teams for pushing the boundaries of what's possible with LeanMCP!

Project Links

• WashedMCP: https://devpost.com/software/washedmcp
• neoCortex: https://devpost.com/software/neocortex
• Checkr: https://devpost.com/software/checkr-esikfo

---

FAQ

Q: What made these projects stand out in the LeanMCP track? A: All three projects achieved significant token savings (70-85%) while solving real-world problems in code intelligence and healthcare, demonstrating practical LeanMCP applications.

Q: How did WashedMCP achieve 70% token savings? A: By using custom TOON (Token-Optimized Object Notation) format and returning callers, callees, and context in a single query instead of multiple separate requests.

Q: What makes neoCortex different from regular code search? A: neoCortex builds bidirectional dependency graphs and uses hybrid search combining embeddings, keywords, and graph traversal for more intelligent results.

Q: How does Checkr maintain context across text and voice interactions? A: Through LeanMCP's shared context capabilities and LiveKit agentic workflows, enabling seamless transitions between communication modalities.

Q: Can I use these projects in my own applications? A: neoCortex is open source on GitHub. For WashedMCP and Checkr, check their Devpost pages for availability and contact information.

Q: What other hackathons feature LeanMCP tracks? A: LeanMCP has supported 6 global hackathons. Check our community announcements for upcoming events and opportunities.

Q: How do I get started building with LeanMCP? A: Start with our SDK documentation and examples. The CLI tool makes it easy to scaffold new projects and understand the patterns.

Q: What domains work well with LeanMCP? A: These winners show LeanMCP works across developer tools, healthcare AI, and any domain requiring efficient tool calling and context management.

Related Resources

• LeanMCP SDK Documentation - Complete development guide and examples
• MCP Production Deployment - How to deploy MCP applications
• Agent Runtime Guide - Parallel MCP execution for performance
• LeanMCP vs Alternatives - Understanding MCP advantages

Join the Community

Ready to build the next breakthrough LeanMCP application?

• GitHub: LeanMCP SDK
• Documentation: leanmcp.com
• Community: Join our developer community pushing the boundaries of agentic AI

Congratulations again to all NexHacks participants and winners!