Claude Code Just Got Updated with One of the Most-Requested User Features

Why It Matters

By freeing the context window, developers can attach thousands of tools without sacrificing performance, accelerating the adoption of agentic AI platforms and reducing operational costs.

Key Takeaways

• Tool Search lazily loads tool definitions on demand.
• Token consumption drops up to 85%, saving context.
• Model accuracy improves 15‑9 points on Opus benchmarks.
• Developers can expose thousands of tools without context penalty.
• Update aligns AI agents with modern software lazy‑loading practices.

Pulse Analysis

The Model Context Protocol (MCP) was introduced to standardize how AI models interact with external tools, but its early implementation forced Claude Code to ingest full documentation for every registered utility. With a 200,000‑token window, developers often sacrificed a third of that capacity merely to load tool metadata, limiting the richness of prompts and inflating compute costs. This context bloat became a bottleneck as ecosystems grew, prompting calls for a more efficient architecture.

Anthropic's MCP Tool Search tackles the problem by adopting a lazy‑loading strategy familiar to modern IDEs. Instead of dumping all tool definitions into the prompt, Claude Code monitors token usage and, once a 10% threshold is reached, swaps the raw docs for a lightweight search index. When a user requests a specific action, the system retrieves only the relevant definition, slashing token usage from roughly 134,000 to 5,000 in internal tests—an 85% reduction. The streamlined context also sharpens the model's attention, boosting Opus 4 accuracy from 49% to 74% and Opus 4.5 from 79.5% to 88.1%.

The broader impact is a shift from a scarcity‑driven "context economy" to an access‑driven model. Developers can now expose thousands of connectors, APIs, and scripts without fearing token penalties, unlocking richer, more capable AI agents. This architectural maturity positions Claude Code as a scalable platform for enterprise automation, lowers operational expenses, and sets a new benchmark for AI‑tool integration standards across the industry.