I have tried Sonnet, Opus and GPT. None of them work.
My Proxy is operating well,not its problem
For AI issues: add Request ID with privacy disabled
Request ID: f374f123-0112-408b-9159-97b9f2c4dc9a
{“error”:“ERROR_BAD_REQUEST”,“details”:{“title”:“Bad Request”,“detail”:“Append data exceeds maximum size of 52428800 bytes”,“isRetryable”:false,“additionalInfo”:{},“buttons”:,“planChoices”:},“isExpected”:true}
Bad Request Append data exceeds maximum size of 52428800 bytes
sBi: Bad Request Append data exceeds maximum size of 52428800 bytes
at Bnw (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:43963:24394)
at Inw (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:43963:23385)
at Onw (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:43964:6355)
at b9u.run (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:43964:11154)
at async qIn.runAgentLoop (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:56307:11753)
at async E0d.streamFromAgentBackend (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:56377:11057)
at async E0d.getAgentStreamResponse (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:56377:17161)
at async o3e.submitChatMaybeAbortCurrent (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:44075:19892)
at async Hl (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:55360:4887)
For AI issues: which model did you use?
Model name (e.g., Sonnet 4, Tab…)
For AI issues: add Request ID with privacy disabled
Request ID: f9a7046a-279b-47e5-ab48-6e8dc12daba1
For Background Agent issues, also post the ID: bc-…
Additional Information
Add any other context about the problem here.
Does this stop you from using Cursor?
Yes - Cursor is unusable
Sometimes - I can sometimes use Cursor
No - Cursor works, but with this issue
The more details you provide, the easier it is for us to reproduce and fix the issue. Thanks!
Hey, thanks for the report. This error usually shows up when the request to the server goes over the 50 MB limit. The two most common causes are:
HTTP/2 is disabled
If you have the Disable HTTP/2 setting enabled (Settings > search “HTTP/2”), the data gets encoded less efficiently and ends up about twice as large. Please check that setting, and if it’s on, try turning it off.
Large files in the project
A 93 MB folder isn’t small. If it contains node_modules, binaries, or other big files, they can get pulled into the context. Try this:
Create a .cursorignore file in the project root and add heavy directories there, similar to .gitignore
For a quick test, open an empty folder and send “Hey”. If that works, the issue is in your project contents.
Also, please check if you have any Skills set in Cursor Settings. You said you don’t, but just to be safe, this is the most common reason for this error: Appendd data exceeds maximum size of 52428800 bytes
I can see the screenshot with the settings. A couple things are worth checking.
Main suspects:
HTTP/1 path
This is the key factor. When Disable HTTP/2 is on, data is encoded less efficiently and the request size can effectively double. You said you can’t use any models except Auto without this. Can you share what exactly happens with HTTP/2 enabled? What error do you see? This might be a separate issue to fix, because HTTP/1 by itself can trigger the limit error.
Binary files in the project
In the sidebar I can see PDF files, .docx documents, and an images folder with large files. Your .cursorignore doesn’t exclude them. Add:
*.pdf
*.docx
*.doc
images/
Quick test
Open a completely empty folder and send “Hey”. If that works, the problem is definitely caused by something in the project content.
Let me know the results, especially about HTTP/2. That’s probably the key to fixing it.
Now about “Model not available”. This happens because Cursor detects your region by IP on the server side. If your VPN exit IP is detected as China, OpenAI and Anthropic models are blocked. That creates a loop. Without a VPN you can’t use the model due to HTTP/2 issues, with a VPN on HTTP/1.1 you hit the size error.
The fix is to set up your VPN so traffic to Cursor servers goes direct without the proxy, and everything else keeps going through the VPN as usual. Depending on your client Clash, V2Ray, Shadowsocks, etc, add these to your bypass list or direct rules:
If your VPN exit IP is not China and it’s still blocked, run Run Diagnostics in Cursor Settings > Network and send the result here. I’m especially interested in the IP: line in the SSL section. If it shows 127.0.0.1, your traffic is still going through a local proxy, and Cursor is seeing the proxy exit IP, not your real IP.
Hey, I can see a screenshot with a Clash config. There are two issues:
Cursor traffic is going DIRECT, and that’s what causes the block
In my previous post, I gave the wrong advice. For China, routing directly DIRECT means Cursor sees a China IP and blocks the models. You need the opposite, route Cursor traffic through a proxy that exits in a supported region like the US, Japan, or the EU.
SELECT_NODE is your proxy group that uses a node in a supported region.
DOMAIN-SUFFIX rules don’t need *
Right now you have DOMAIN-SUFFIX,*.cursor.sh,DIRECT. The * is unnecessary. DOMAIN-SUFFIX,cursor.sh already matches all subdomains like api.cursor.sh, www.cursor.sh, etc.
After the changes:
Make sure the selected node exits in the US, Japan, or the EU
Turn HTTP/2 back on, uncheck Disable HTTP/2
Restart Cursor
Try Sonnet 4.6
Let me know how it goes, and sorry again for the confusion in my previous advice.
