Skip to content

Implement tool_search and code execution by AI agent as MCP client #5

@timasin

Description

@timasin

Anthropic declared the early discovery and loading of all tools from any mcp server causes context bloating which leads to less tokens available for actual tasks that needs to be done on top of the mcp tools.
Thats why they introduced tool_search tool that helps discover only the needed tools for a given prompt. The idea of tool discovery is not new and helps in communication overload in any system which can afford to delay overload of data.
They also propose to add code snippets that can be executed by Agents to execute and extract the result from multiple tools and the result should be sent to the LLM instead of the bloated context of tools and its responses.

We need to implement a functionality to defer tool loading and allow tool search as well.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request
    No fields configured for Feature.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions