Early Preview: Unified MCP Tool Graph – Feedback, Testing, and Collaboration Welcome #7
pratikjadhav2726
started this conversation in
Ideas
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Hi everyone,
I’ve opened up the first version of the Unified MCP Tool Graph –
a project aimed at solving a major bottleneck in agentic systems:
how to dynamically retrieve and select the right tools across thousands of APIs and MCP servers — without hardcoding or manual setup.
Right now, the project includes:
• A Neo4j-powered tool graph with vector search
• Dynamic retrieval based on semantic matching of tool descriptions
• Retrieval optimized for available user credentials (you only get tools you can actually use)
• A FastMCP server ready for LangChain / LangGraph agent integration
⸻
🎯 What works today:
• Load tools from multiple MCP servers
• Embed tool descriptions locally
• Query the graph with task descriptions
• Return a minimal set of relevant tools for any task
• Run a FastMCP retrieval server on top of the graph
⸻
📋 Areas still under active development:
• Improving retrieval scoring and fallback strategies
• Adding relationships like overlaps_with, preferred_for_task
• Extending to larger multi-domain toolsets
• Streamlining deployment for production use
• Cleaner client interfaces and LangChain templates
⸻
🚀 I’d love your help with:
• Testing retrieval with different types of tasks and agents
• Finding edge cases or ideas for smarter retrieval
• Suggestions for better metadata organization or graph expansion
• Collaboration on scaling this beyond a prototype
Beta Was this translation helpful? Give feedback.
All reactions