One-line instruction beats long tool descriptions: MCP eval hits 18/20 vs 3/10
colinmcnamara · x · 2026-10-23
At @AustinLangChain, Jeff Linwood shared an eval for anyone shipping MCP tools: a single short instruction got agents to use his task tracker in 18 of 20 runs, while relying on long tool descriptions yielded 3 of 10 or less.
The takeaway: write the one line. For MCP developers, a concise directive beats verbose tool descriptions at a fraction of the effort.
Related event: One-line instruction dramatically boosts agent's MCP tool usage(2 posts)→
More from coding & agent
- AutoRouter launches beta: auto-routes each coding task across Claude, GPT and other models to cut token spend — floguo · 2026-10-23
- How Speak uses GPT-Live and a Decisions API for ultra-responsive live AI language tutoring — pbbakkum · 2026-10-23
- Google Live API adds decisions model triggers, faster than function calls for voice agents — pbbakkum · 2026-10-23
- An agent left a note in a package folder — and another agent found it — petrusenko_max · 2026-10-23
- Daily Codex + Claude Code user: session bookkeeping, not code, is the bottleneck — KR4K3N1S · 2026-10-23
- Claude Code's bundled Bun 1.4.3 ships RCE flaw; update past 2.1.308 — KyeGomezB · 2026-10-23