[Enhancement] Avoid MCP Client from Re-interpreting specialised tool result #75

Currently, certain tools are specially formatted by the backend but when transferred, the MCP client might re-interpret the results, causing a loss in nuance of unfaithful representation of the actual output. We need separate branching logic on whether to display the output verbatim, or allow the LLM / MCP client to re-interpret.

In both cases, the MCP client should still be aware of the full tool output and store it as part of conversation history or context. This avoids seeming disconnect with users' follow-up questions.