Where does the bug appear (feature/product)?
Cursor IDE
Describe the Bug
I am experiencing a technical failure where the agent is unable to execute image generation, even though the model’s internal reasoning logs show it is explicitly attempting to do so.
The Setup:
- Cursor Version: 2.4.21
The Problem:
When I prompt the agent to generate an image (e.g., “Create a high-fidelity logo for this landing page”), the agent understands the request perfectly. In the “Thinking” / Reasoning trace, I can see the model explicitly deciding to call the image generation tool:
Example from logs: “I will now use the image generation tool to create the requested asset using the Nano Banana model…”
The Failure:
Despite the model emitting the intent to call the MCP tool/image generation service:
-
No image preview appears.
-
No file is created in the assets/ folder.
-
The agent eventually “hangs” or responds with text as if it performed the action, but no tool output is actually generated.
My Analysis:
Since the model wants to call the tool but the IDE doesn’t execute it, this feels like a permissions or billing-gate issue.
I am on the Legacy Pro plan (500 requests). It appears that the Agent Harness in the latest build (2.4+) sees the request for a high-cost asset (Nano Banana) and silently fails the MCP call because it cannot deduct the cost from a “Request-based” counter (as opposed to the new Credit-based system).
Questions for the Engineering Team:
-
Why is the MCP call for image generation failing silently instead of providing an error message?
-
Is image generation explicitly disabled for Legacy Pro users, even if the model’s reasoning suggests it should be available?
-
If this is a billing restriction, can you implement a proper UI notification (e.g., “Upgrade to Credit-based billing to use this tool”) rather than a silent failure in the Reasoning chain?
I would appreciate it if someone from the dev team could confirm if the “Legacy 500-request” backend is technically capable of executing the Nano Banana MCP calls.
Thank you!
Steps to Reproduce
When I prompt the agent to generate an image (e.g., “Create a high-fidelity logo for this landing page”), the agent understands the request perfectly. In the “Thinking” / Reasoning trace, I can see the model explicitly deciding to call the image generation tool:
Example from logs: “I will now use the image generation tool to create the requested asset using the Nano Banana model…”
Operating System
Windows 10/11
Version Information
Version: 2.4.21 (system setup)
VSCode Version: 1.105.1
Commit: dc8361355d709f306d5159635a677a571b277bc0
Date: 2026-01-22T16:57:59.675Z
Build Type: Stable
Release Track: Nightly
Electron: 39.2.7
Chromium: 142.0.7444.235
Node.js: 22.21.1
V8: 14.2.231.21-electron.0
OS: Windows_NT x64 10.0.26200
Does this stop you from using Cursor
Sometimes - I can sometimes use Cursor