OpenRouter measures agent tool context costs and offers Tool Search to reduce them
Overview
OpenRouter's blog measures how MCP servers, CLIs, and skills consume agent input tokens.
On one GitHub task, the default 46-tool GitHub MCP server used 59,417 input tokens on Claude Sonnet 5.5, compared with 1,421 for the gh CLI. The post also explains how to count tool-definition tokens and how OpenRouter's Tool Search server tool reduces up-front cost. OpenRouter says the results vary by model.
Tool Search is in beta and works with any model and provider on the Responses and Messages APIs. The Chat Completions API rejects openrouter:tool_search with a 400 error, so agents on that API must move to the Responses or Messages API first.
Written by AI from the articles below · updated Nov 10, 8:11 PM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- OpenRouter BlogOfficialPickOpenRouter measures how MCP, CLI, and skills cost agent context tokens
AIOpenRouter measured input tokens for one GitHub task across a CLI, a skill, and the GitHub MCP server on several models. The default 46-tool MCP server used 59,417 input tokens on Claude Sonnet 5.5, against 1,421 for the gh CLI. The post shows how to count tool-definition tokens and how its Tool Search server tool reduces up-front cost, with results that vary by model.
Heat Trend
- Current comparable heat
- 7
- Peak comparable heat
- 9Nov 10, 9:00 PM ET
- Change in past 24 hours
- –
The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.