Loading page...Загружаем страницу...正在加载页面...
AtlasRepo

What AI agents actually do with your MCP server

Описание и применение

Серверы MCP могут казаться здоровыми, но агенты сталкиваются с проблемами из‑за отсутствующих инструментов, неверных аргументов, некорректных описаний, повторяющихся вызовов и скрытых размеров ответов. Инструмент mcpspan выявил эти паттерны, позволяя улучшить взаимодействие без изменения кода.

Задача

MCP servers can appear healthy yet agents struggle due to missing tools, bad arguments, misleading descriptions, looping calls, and hidden response size issues.

Результат

Instrumentation revealed five key failure modes: missing tools, argument mismatches, misleading descriptions, looping calls, and hidden response size spikes. Addressing these improved agent reliability without code changes.

Рабочий процесс

  • Instrument server with mcpspan
  • Collect agent request logs
  • Identify missing tool calls
  • Detect schema‑rejection errors
  • Analyze tool description mismatches
  • Track repeated calls per session
  • Measure response sizes per tool
  • Adjust tool names, descriptions, and limits
  • Re‑instrument to verify improvements

Доказательства

Agents guess tool names. Some guesses are near misses: searchFlights when the tool is search_flights, seatMap instead of seat_map. Others are a tool the agent assumed should exist, like get_flight_price.

Источник

When an agent sends arguments that don't match the input schema, the MCP SDK rejects them before your handler runs and sends the validation error back to the model as a normal result.

Источник

book_flight took passengers, described as "the passengers". Agents read that as a count and sent 2. The schema wanted a list of names, so the SDK refused the call.

Источник

An agent tries check_in, gets "Check-in opens 24 hours before departure", and calls check_in again with exactly the same arguments. Then again.

Источник

A tool that usually answers with 6 kB and sometimes with 650 kB looks fine on every latency chart, because it's fast. But that answer lands in the agent's context window, crowds out everything else, and costs tokens on every following turn.

Источник

  • Дата записи: 2026-10-21T14:56:27.901Z