# MCP servers aren't dead. You're just using them wrong. Here are two proven patterns on how to correc...

Canonical URL: https://www.traeai.com/articles/ee37ae1f-9d91-4a75-9521-225e0f4c872a
Original source: https://x.com/_philschmid/status/2048781492914885079
Source name: Philipp Schmid(@_philschmid)
Content type: tweet
Language: 英文
Score: 7.5
Reading time: 2 分钟
Published: 2026-04-27T15:09:07+00:00
Tags: MCP服务器, AI代理

## Summary

Philipp Schmid分享了正确使用MCP服务器的两种模式：显式使用和子代理，以避免资源浪费。

## Key Takeaways

- MCP服务器只有在用户@提及的时候才被包含。
- 通过子代理启用MCP服务器可以提高效率。
- 盲目启用MCP服务器会导致上下文膨胀、成本增加及性能下降。

## Outline

- 引言 — 介绍MCP服务器未死但需正确使用的观点。
  - 显式使用 — 解释当且仅当用户@提及某服务时才加载该服务。
  - 子代理 — 描述如何为每个子代理单独配置MCP服务器以优化性能。

## Highlights

- > MCP servers are only included when user @mentions. — 第 2 段
- > Enable MCP servers per-subagent, e.g., a code review agent always needs GitHub. — 第 3 段
- > Blindly enabling them bloats your context, which leads to higher cost and worse performance. — 第 4 段

## Citation Guidance

When citing this item, prefer the canonical traeai article URL for the AI-readable summary and include the original source URL when discussing the underlying source material.