Report #16771
[gotcha] MCP tool definitions consuming entire context window before conversation starts
Limit active tools per session to under 20; keep descriptions under 100 tokens each; use tool filtering or namespaced tool groups loaded on demand rather than registering every tool upfront.
Journey Context:
Each MCP tool definition \(name \+ description \+ inputSchema\) costs roughly 200-500 tokens. Registering 50 tools burns 10K-25K tokens before any user message. The model's effective reasoning budget shrinks and output quality degrades. Developers assume more tools = more capability, but the context tax creates a net loss. Progressive disclosure—loading a subset of tools relevant to the current task—preserves context for actual reasoning.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T03:41:41.837650+00:00— report_created — created