How Claude Code's Tool Search Saves 90% of Your Context Window

Claude Code's tool search feature defers loading of MCP tool definitions, reducing token overhead and preserving context window for actual work.

💡

Why it matters

This feature represents a fundamental shift in how Claude Code handles tool ecosystems, reducing token overhead and preserving context window.

Key Points

  • 1Tool search replaces multiple MCP tool definitions with a single search tool that loads tools on-demand
  • 2This reduces token overhead from ~90,000 to ~15,000 tokens immediately
  • 3Tool search operates in three modes to balance savings and latency
  • 4Discovered tools are preserved across context compaction through a snapshot system

Details

Claude Code's tool search feature automatically defers loading of MCP tool definitions, replacing them with a single search tool that loads tools on-demand. This reduces the token overhead from ~90,000 to ~15,000 tokens immediately, preserving the context window for actual work. The system uses a priority-ordered checklist to determine which tools to defer, with the goal of 'fail closed, fail toward asking' - if anything is uncertain, the system loads all tools rather than hiding them. Tool search operates in three modes - the default 'tst' mode always defers MCP and 'shouldDefer' tools, 'tst-auto' mode defers based on a token budget threshold, and 'standard' mode never defers. Discovered tools are preserved across context compaction through a snapshot system, ensuring Claude doesn't lose access to tools it's already discovered.

Like
Save
Read original
Cached
Comments
?

No comments yet

Be the first to comment

AI Curator - Daily AI News Curation

AI Curator

Your AI news assistant

Ask me anything about AI

I can help you understand AI news, trends, and technologies