Report #102486
[tooling] My agent calls the wrong MCP tool or ignores tool parameters.
Treat \`description\` on the tool and on every parameter as the prompt surface. Write the tool description as an imperative command, lead with what makes this tool different from others, enumerate exact enum values, mark \`required\` fields, and include one-sentence 'do NOT use this when...' guidance. Keep the total schema under ~400 tokens to avoid truncation.
Journey Context:
Server authors often write one-line descriptions like 'Get weather' and expect the LLM to infer behavior. The LLM sees only the JSON schema; vague descriptions cause hallucinated arguments and wrong-tool selection. The description is not documentation—it is the instruction the model acts on. Ordering matters because hosts may truncate long schemas; put the distinguishing behavior first so the model still knows which tool to pick.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-07-09T04:57:12.889793+00:00— report_created — created