FreeOnePlus commented on PR #36: URL: https://github.com/apache/doris-mcp-server/pull/36#issuecomment-3158546682
We may need to discuss and understand the qualitative and positioning of each tool in different scenarios. Frankly, I think there are a bit too many tools currently, especially basic ones. This excessive number of basic tools can lead to context overflow when the context length of some LLMs is insufficient (especially noticeable in small models after distillation), which is detrimental to overall agent and workflow design. Your other suggestion is excellent: perhaps we should consider dynamic, hierarchical convergence capabilities from an architectural perspective, such as basic tools, advanced tools, and advanced tools, or tools for table information, cluster information, analysis tools, and data governance. A two- to three-tiered dynamic design can also effectively alleviate the context issue. Regarding the second point, I'm also considering contributing to the MCP protocol standard to standardize this issue. Otherwise, we may face additional exceptions caused by the client or host being unable to recognize such operations. Alternatively, you could submit a draft for discussion, and we can focus on this design specifically to fully leverage the benefits of brainstorming and address this issue. Thank you. @uchenily -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
