[FEATURE]

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Raw MCP results (especially full file reads) are token-dense garbage that burns context fast. It's doubly wasteful: costs tokens to inject, then costs again when the model
processes them. That's the whole reason ROME-FIRST exists — SAFE_SHELL and GEMINI handle the heavy I/O off-context, and only summaries land here.

Proposed Solution

A flag like --no-tool-results-in-context or scoping it per MCP server would let paid-model sessions stay lean. ?

Alternative Solutions

_No response_

Priority

Critical - Blocking my work

Feature Category

CLI commands and flags

Use Case Example

_No response_

Additional Context

_No response_