@alex_prompter: My agents kept getting dumber every time I gave them more tools. The reason is mechanical. Every MCP server you connect…

X AI KOLs Timeline Tools
ai-agent tool-selection context-efficiency open-source bm25

Summary

Ratel is an open-source tool that reduces input tokens by 79% and improves tool selection accuracy for AI agents by loading only needed tools using a BM25 index, instead of all available tools.

My agents kept getting dumber every time I gave them more tools. The reason is mechanical. Every MCP server you connect dumps its full tool list into context on every turn, and the right tool ends up buried under 100 descriptions the model will never use. Ratel flips it: it sits between your agent and the catalog and loads only what the turn needs. No embeddings, no vector database, a plain BM25 index in process. Their bench: 79% fewer input tokens on Sonnet at 180 tools. Local models jump from 8% to 77% tool selection accuracy. The repo is open. Context, not capability, is the main issue for agents right now.
Original Article
View Cached Full Text

Cached at: 07/22/26, 08:28 AM

My agents kept getting dumber every time I gave them more tools.

The reason is mechanical.

Every MCP server you connect dumps its full tool list into context on every turn, and the right tool ends up buried under 100 descriptions the model will never use.

Ratel flips it: it sits between your agent and the catalog and loads only what the turn needs.

No embeddings, no vector database, a plain BM25 index in process.

Their bench: 79% fewer input tokens on Sonnet at 180 tools.

Local models jump from 8% to 77% tool selection accuracy.

The repo is open.

Context, not capability, is the main issue for agents right now.

Ratel (@ratelai): We just made agents 7x cheaper to run

Today we’re releasing Ratel

Right now your agent loads all the tools and skills you own to do one small task Ratel loads only what that task needs

See how it works ↓

Similar Articles

@akshay_pachaar: https://x.com/akshay_pachaar/status/2053166970166772052

X AI KOLs Timeline

The article discusses a shift in AI agent tool usage from the 'MCP vs CLI' debate to 'Code Mode,' where agents write code to dynamically import tools, significantly reducing context window usage. It highlights Anthropic's approach and Cloudflare's implementation, demonstrating a 98.7% reduction in token consumption for specific tasks.

Writing effective tools for agents — with agents

Anthropic Engineering

Anthropic shares engineering best practices for designing, evaluating, and optimizing tools for AI agents, specifically utilizing the Model Context Protocol (MCP) and Claude Code to improve agent performance.