MCP tool input schemas are stripped of type/enum/default — some properties sent as empty {}

When forwarding MCP tools to Claude/Anthropic, BoltAI appears to apply a lossy transformation to each tool's input_schema, dropping type information that the MCP server provides:

• Several properties are emitted as empty schemas with no type, e.g. "query": {}, "continuation": {}, "channel": {}, "playlist": {}.

• Properties that have enum/default in the source schema lose them — e.g. format is sent as {"type": "string"} instead of {"type": "string", "enum": ["json","text"], "default": "text"}.

This is valid JSON Schema ({} matches any type) and is accepted by the API, so it's not a hard error — but it's a quality regression. With no type, the model has to guess the expected argument type, which reduces tool-call reliability (malformed or wrong-typed arguments). Dropped enum/default values remove guidance the model would otherwise use.

Confirmed by comparison: another client (not named) hitting the same MCP server forwards the full schemas intact — typed properties plus enum and default values — while BoltAI's version is stripped down. This points to a transformation in BoltAI's MCP→Anthropic schema handling rather than the MCP server's output.

Request: forward the MCP server's tool inputSchema unchanged — preserve each property's type, enum, and default.