Show HN: An MCP proxy that sends 97% fewer tool-definition tokens per request (github.com)

The recent launch of the mcp-toolsearch-proxy introduces a significant advancement in how AI models access and utilize multiple tools simultaneously. By allowing models to engage hundreds or thousands of MCP tools without the burden of loading extensive tool-definition tokens, it reduces the token usage per request by an impressive 97%. In practical terms, this means that a model can efficiently work with around 500 tools while incurring only 1,735 tokens for tool definitions, significantly lowering operational costs and maintaining high accuracy—achieving 90% success in identifying the correct tool compared to 95% when all tools are loaded. This innovation is particularly valuable for the AI/ML community as it streamlines the querying process, enabling models to call and utilize tools more judiciously and cost-effectively. The mcp-toolsearch-proxy features an advanced search mechanism that combines lexical matching with multilingual embedding models, allowing for rapid, efficient tool retrieval across multiple languages without the need for heavy computational resources. This not only enhances accessibility for diverse global users but also establishes a scalable framework that can evolve as the number of available tools increases. Overall, this development encourages more efficient tool integration in AI applications, offering a robust solution to manage the complexities of tool-heavy environments.