State-Aware Dependency System Help the tool

[whatevertogo]

Allow AI to search for the content it needs through multi-level tools layer by layer, rather than giving it everything at

[Scriptwonder]

I am actually working on this at the same time! Sounds like something is needed before Anthropic throws another cool trick at us.
I don't think our tool layer is really comprehensive at this point in time, though. We can review code changes, but make sure to have a comparison on tokens/performances so we can better evaluate.

[whatevertogo]

I am actually working on this at the same time! Sounds like something is needed before Anthropic throws another cool trick at us. I don't think our tool layer is really comprehensive at this point in time, though. We can review code changes, but make sure to have a comparison on tokens/performances so we can better evaluate.

Actually, I want to consider solving the problem from the mcp architecture (FaseMCP) layer, rather than through something like Custom_Tool call. However, I'm concerned that this might require a lot of time to implement.

[Scriptwonder]

Then I guess it is outside the current repo usage (or its just on the python side)? Also, does Issue#560 kind of serve as a similar way to deal with optimization?

[whatevertogo]

just the python side i think

[whatevertogo]

Then I guess it is outside the current repo usage (or its just on the python side)? Also, does Issue#560 kind of serve as a similar way to deal with optimization?

Implement a state-aware dependency system for tool management. The C# side should expose an HTTP API (via EditorStateProvider) to report real-time editor states (e.g., compiling, playing, selection). The Python MCP Server will then use this state to dynamically filter the tool list via a prerequisite_check decorator, ensuring AI agents only see actions that are currently valid.

Control tool visibility based on real-time environmental states (e.g., whether objects exist ) rather than on historical steps. Tools are coexistent rather than mutually exclusive, allowing free jumping between different levels.

That's my idea,so what do you think

[Scriptwonder]

Just saw this. I still don't see the exact usage of this, as far as this seems to be a way to build the LLM's understanding without relying on previous history (which isn't realistic, considering this is LLM); As towards an HTTP api, how many tokens will it burn just for the MCP server to retrieve the editor state at a certain frame?
And for the end goal, can you let me know which tools NEED to be filtered at certain times for editor states?