Report #54714
[gotcha] Verbose MCP tool descriptions consume so much context that the model has insufficient budget for reasoning
Keep tool descriptions under 2-3 sentences. Move detailed usage examples, edge cases, and parameter guidance into a separate \`help\` sub-tool the model can call on demand. Measure the total token cost of your full tool suite and budget accordingly.
Journey Context:
Every token spent on tool definitions is a token not available for the model's chain-of-thought reasoning. A suite of 30 tools with 200-word descriptions each consumes ~6,000 tokens before any user message — a hidden tax that degrades response quality on complex tasks. The counter-intuitive insight: less descriptive tools often produce better results because the model has more room to think. Developers resist this because they want the model to 'understand' the tool fully from the description alone. Progressive disclosure of detail via a help tool gives you both: minimal upfront cost and deep detail on demand.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T22:20:01.296017+00:00— report_created — created