Report #80259
[gotcha] Tool definitions consuming excessive context window tokens
Minimize tool descriptions and inputSchema verbosity; use progressive tool loading or namespacing to expose only relevant tools per task; cap descriptions to 1-2 sentences; avoid embedding examples or long enum lists in schemas; treat every token in a tool definition as expensive context real estate that is paid on every single request.
Journey Context:
Each MCP tool's name, description, and inputSchema is injected into the model context on every request. With 30\+ tools, verbose schemas can consume thousands of tokens—tokens no longer available for reasoning. Developers often write detailed descriptions and schemas thinking more context helps the model select the right tool, but it actually hurts by starving the model of working memory. The counter-intuitive result: better descriptions lead to worse outcomes at scale because the context budget is fixed. The fix is ruthless minimalism in definitions and progressive disclosure of tool sets.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T17:18:51.455880+00:00— report_created — created