Feature request: Semantic search for the dynamic tool filtering

[qdrddr]

More than 50 tools can significantly decrease the quality of the response and confuse LLM. Assuming each MCP server holds 10 tools on average, you end up with only 5 MCP servers.

We need to filter out available tools across all the MCP servers dynamically. One way to do that is to ingest descriptions of tools in a vector database such as LanseDB and perform a semantic similarity search to extract only semantically close. This provides a filtered and limited list of tools relevant to the user's question.

[qdrddr]

Here is what I propose for the MCP specification update to facilitate dynamic tool discovery. Please support the idea here

modelcontextprotocol/modelcontextprotocol#845

[qdrddr]

Here is a paper that may be relevant

[qdrddr]

I think this is relevant.

Addressing LLMs limitations in generating sophisticated long-form outputs. Survey analysis of over 1400 research papers:

• Context Retrieval and Generation
• Context Processing
• Context Management
• Memory Systems
• RAG
• Tool-Integrated Reasoning

https://github.com/Meirtz/Awesome-Context-Engineering
https://arxiv.org/abs/2505.06416
https://arxiv.org/abs/2505.03275

[qdrddr]

Context Rot Affects LLM Performance

Longer input does not guarantee consistent results

🔍 Chroma researchers tested 18 LLMs on simple tasks
📉 Found performance declines with longer inputs

• 📏 Input length caused unexpected reliability issues
• 🧪 Highlights the need for long-context evaluations
• 🧠 Suggests better context engineering strategies

Research results https://github.com/chroma-core/context-rot

[qdrddr]

TypeScript Implementation of MCP Tool Semantic Search
https://github.com/samanhappy/mcphub

[qdrddr]

Reranker with graphs
https://github.com/Bavalpreet/MediumBlogs/blob/main/Knowledgegraph%20Reranking/Knowledge_graph_re_ranking.ipynb

[qdrddr]

Suggested Embedding model: Codestral-Embed from mistral and Mxbai-rerank-v2 to improve performance for, code, MCP, and tool retrieval.

[qdrddr]

Suggested Embedded (In-Progress) VectorDBs:

https://github.com/tursodatabase/turso
https://github.com/chroma-core/chroma
https://github.com/lancedb/lancedb
DuckDB with VSS extention

[qdrddr]

Would appreciate if you vote for this improvement idea to standardize MCP Specification with Delegated Advanced Tool Search in my comment feature here.

[BradKML]

Seconding this, but what are the minimal code change needed to enhance this repo?

[qdrddr]

MCP-use lib implemented a tool semantic search mechanism into MCP Client
https://github.com/mcp-use/mcp-use

[qdrddr]

MCP-Agent with EmbeddingRouter for semantic tool search and filtering.

https://github.com/lastmile-ai/mcp-agent

[qdrddr]

MCP-Universe: Benchmarking Large Language Models with
Real-World Model Context Protocol Servers.

Key findings:
“Long-Context Challenge

Token count increases rapidly with interaction steps, often leading to context overflow and degraded performance in multi-step tasks requiring extensive reasoning.”

This proves how MCP tool pre-filtering (prior LLM) is important, and as I was saying, especially manifests in multi-steps.

https://mcp-universe.github.io