Open
Description
As the MCP server continues to expand its integration with various Grafana products and services, the number of available tools exposed to the LLM has grown substantially. With this expansion across Prometheus, Loki, Grafana Dashboards, OnCall, and other services, the LLM faces an increasingly complex decision when selecting the right tool for each query. We need a smarter tool filtering system to maintain performance and accuracy as we continue to scale our support for the Grafana ecosystem.
We need to add a way to help with:
- Categorization system for all tools (Prometheus, Loki, OnCall, etc.)
- Dynamic tool filtering to only expose the relevant subset of tools to the LLM
- Description enhancements for better tool selection
This solution addresses several important benefits:
- Reduced token usage: Only sending relevant tool descriptions to the model
- Improved accuracy: The model is less likely to select inappropriate tools
- Faster responses: Less context for the model to process
- Better user experience: More relevant and precise tool selection