Feature request for product/service
Cursor IDE
Describe the request
The Problem:
I use Cursor’s Agents Window extensively with external tools like ha-mcp. To prevent unexpected state changes or dangerous write operations, I maintain a strict manual approval policy for all MCP write operations.
However, this creates significant friction when executing complex, multi-step plans. Once I have manually reviewed and approved a generated plan document, I have effectively authorized the downstream operations. Forcing manual approval for every sequential write operation during execution interrupts the workflow and defeats the purpose of autonomous plan execution.
Proposed Solution:
Introduce a contextual “Auto-Approve” mode for MCP commands that triggers under specific conditions:
- The agent is actively implementing a previously developed and user-approved plan document.
- A lightweight subagent evaluates the requested MCP operation and confirms it is strictly necessary to complete the current step of the plan.
Potential Implementation Detail:
If having a subagent evaluate intent against a free-text plan is too error-prone or slow, the initial plan-generation workflow could be modified. The planner could explicitly output an array/manifest of required MCP operations alongside the text. The execution subagent would then simply cross-reference the requested tool against this pre-approved manifest.