How the design of an MCP server affects agent token usage

Description and use

Adding composite operations to the MCP reduced the number of calls and tokens required by agents to edit Word documents, improving efficiency without compromising correctness.

Problem

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

Outcome

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

Workflow

• 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

Evidence

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

Source

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

Source