[FEATURE]: Claude Advanced Tool Usage

[TheKyleChau]

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Anthropic released new advanced tool usage for the Claude API which reduces amount of tokens spent by pre-loading tools into context and only needing them when necessary. It's current in beta but I think this could be very beneficial to those who use Claude models frequently and I think is a good feature to add, especially Programmatic Tool Calling which has Claude write python code to create it's own tool calling.

Today, we're releasing three features that make this possible:

Tool Search Tool, which allows Claude to use search tools to access thousands of tools without consuming its context window
Programmatic Tool Calling, which allows Claude to invoke tools in a code execution environment reducing the impact on the model’s context window
Tool Use Examples, which provides a universal standard for demonstrating how to effectively use a given tool

https://www.anthropic.com/engineering/advanced-tool-use

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• Anthropic API Context Management (context-management-2025-06-27) #2871: Anthropic API Context Management (context-management-2025-06-27) - Related to Anthropic's context management features
• Feature Request: Tool to list and manage MCP servers to reduce context window #2418: Feature Request: Tool to list and manage MCP servers to reduce context window - Related to optimizing context window usage
• [FEATURE]: Context protection for tool calls when response is large #4560: [FEATURE]: Context protection for tool calls when response is large - Related to efficient context usage with tool responses

Feel free to ignore if none of these address your specific case.

[ethan-huo]

anthropic is just a post‑hoc fix for problems that shouldn't exist; we only need to expose the mcp schema as a TypeScript SDK, which can be discovered step by step, and provide code‑execution style calling capability.

For example, mcp.json can be regarded as in the system context
mcp.json should appear in the system prompt as

Available modules:
notion(30 tools, 1000 token), supabase(10 tools, 600 token), github(30 tools, 2000 token)

Use searchTool search tool
search('notion', 'page' )  -> return markdown typescript api declare

list( module )  

// Execute script calling tool
execute( `

const page = notion.get_page('')

const result = supabase.insert( page )

console.dir( reuslt, { depth: null } )
` ) 

[TheKyleChau]

anthropic is just a post‑hoc fix for problems that shouldn't exist; we only need to expose the mcp schema as a TypeScript SDK, which can be discovered step by step, and provide code‑execution style calling capability.

For example, mcp.json can be regarded as in the system context mcp.json should appear in the system prompt as

Available modules:
notion(30 tools, 1000 token), supabase(10 tools, 600 token), github(30 tools, 2000 token)

Use searchTool search tool
search('notion', 'page' )  -> return markdown typescript api declare

list( module )  

// Execute script calling tool
execute( `

const page = notion.get_page('')

const result = supabase.insert( page )

console.dir( reuslt, { depth: null } )
` ) 

That's basically the thing Anthropic is completing but natively within the API? I think this would take even more context due to the extra MCP lines having to be in context.

[ethan-huo]

That's basically the thing Anthropic is completing but natively within the API? I think this would take even more context due to the extra MCP lines having to be in context.

What I mean is that there’s no need to load mcp and related information into the context; you only need to give the model basic information: module, list, search, execute. The model will accomplish the most with the least token consumption. mcp is just an API form.