Right now Boost registers 15 MCP tools. That means ~2,100 tokens of schemas loaded into every single message in the conversation, and the AI typically needs 3-5 sequential tool calls to gather context for a task (e.g. schema, routes, errors, app info one by one).

What if instead of 15 individual tools, there were just 2-3 that let the AI request everything it needs in one call?

Something like:

• A lightweight manifest tool that returns a compact catalog of whats available (~200 tokens instead of ~2,100)
• A resolve tool that accepts multiple context requests at once and returns them batched in a single respone
• Keep execute as-is for code execution

This could also open the door for pre-defined bundles like "give me everything I need for debugging" or "give me database context" so the AI doesn't even have to think about which pieces to request.

From some rough estimates this could cut schema overhead by ~80% and reduce round-trips from 5 to 2 for most tasks.

Happy to start working on this if you think it's worth exploring.