How the design of an MCP server affects agent token usage

说明与用途

在 MCP 中添加复合操作减少了代理编辑 Word 文档所需的调用次数和输出令牌数量，提高了效率且不影响正确性。

问题

MCP servers often require agents to perform many small tool calls, increasing output tokens and protocol overhead when editing Word documents.

结果

Composite MCP operations cut MCP calls by ~29% and reduced output tokens by 4–16% across models, improving efficiency without sacrificing correctness.

工作流

• Identify frequent multi‑call sequences in agent traces
• Create composite operations that combine those calls
• Add the composites to the MCP tool surface without removing primitives
• Run controlled experiments comparing shortcut vs. no‑shortcut surfaces
• Measure MCP calls, output tokens, latency, and correctness
• Analyze aggregate results and remove ineffective shortcuts

证据

Total number of MCP calls fell from 501 to 358—a reduction of approximately 29%

来源

Median of the seven task‑level shortcut/no‑shortcut output‑token ratios was 0.85x to 0.96x

来源