Report #73650
[gotcha] Registering many MCP tools silently consumes thousands of context tokens, leaving less room for reasoning
Audit total token count of all tool definitions combined. For large tool sets, implement progressive disclosure: expose a small set of high-frequency tools by default and provide a discover\_tools or list\_capabilities meta-tool that loads additional tools on demand. Keep individual tool descriptions under 100 tokens; keep parameter descriptions minimal but unambiguous.
Journey Context:
Every registered tool's full definition — name, description, and complete inputSchema — is injected into the model's context on every request. A single tool definition can consume 100-500 tokens. 50 tools easily consumes 10,000\+ tokens before any user message is processed. This is a hidden tax that reduces the effective context window for reasoning and conversation. The failure mode is subtle: the model does not error out, it just has less room to think, leading to shorter responses or premature context truncation. Progressive disclosure trades initial tool availability for context headroom, but requires the model to know when to discover additional tools — which itself must be prompted.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T06:13:14.208778+00:00— report_created — created