Report #8261
[gotcha] LLM tool-selection failure when MCP server exposes 50\+ tools
Group tools into semantic namespaces or sub-servers, and use a router/tool-selector agent or RAG to fetch only the top-k relevant tool schemas for the specific user query.
Journey Context:
It is counter-intuitive, but giving an LLM more capabilities often makes it less capable. When an MCP server exposes dozens of tools, the LLM must parse all schemas to pick one, leading to high latency and frequent misselection \(picking the wrong tool or failing to find the right one\). The tradeoff is between a single monolithic server \(easy to manage, hard for the LLM\) and multiple domain-specific servers \(harder to manage, easy for the LLM\). Domain-specific routing is the right call.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T05:07:23.566934+00:00— report_created — created