Configurable max tool response size

[strawgate]

I have written a similar (not anywhere near as good framework) to MCP agent and I'm currently attempting to switch to MCP agent.

One of the things that was important for me limiting LLM failures was being able to set max tool response sizes in the calling code (as I don't control the MCP servers).

If the LLM makes a call that returns say 50k tokens that might be fine but one that returns 500k tokens would get rejected with an error telling the LLM to use a more specific query/etc.

I think I can add this by inheriting Agent and customizing the call tool method but I wanted to flag it as a feature request here too

[saqadri]

@strawgate loving all the feedback - keep it coming! What do you think of adding it as a property to RequestParams? And having AugnentedLLM enforce it?

[strawgate]

Yeah that sounds like a great place for it.

On a side note, one of the things I'm trying to put together is a "best practices analyzer" that orchestrates (parallel?!) something like an LLM map reduce on a - large volumes of arbitrary data (>100mb+) to provide a "current practices" analysis of a data set (like documentation) that I can then use to guide creation of similar content.

I've got a couple of projects I'll convert this weekend and will follow-up with anything else I notice

[saqadri]

Sweet! Keep me posted how the conversion goes, will prioritize issues you run into. Also, Parallel workflow is great for map-reduce style flows.