Openrouter with qwen/qwen3-32b model with mcp server gives error 400 Input required: specify "prompt" or "messages"

[iliyanpetrovcleverpine]

What happened?
Setup: Openrouter with qwen/qwen3-32b and MCP server https://demo-day.mcp.cloudflare.com/sse
When I ask for "When's MCP Demo Day? Should I get excited?" as prompt I get the block which displays the mcp server response but then an error is returned

Something went wrong. Here's the specific error message we encountered: An error occurred while processing the request: 400 Input required: specify "prompt" or "messages"

The same prompt works with chatgpt-4o without any issues.

Version Information
LibreChat 0.7.8

Steps to Reproduce

• Configure OpenRouter and select Model qwen/qwen3-32b
• Configure mcp server https://demo-day.mcp.cloudflare.com/sse which doesn't require authentication
  Prompt:"When's MCP Demo Day? Should I get excited?"

UI Error:

Something went wrong. Here's the specific error message we encountered: An error occurred while processing the request: 400 Input required: specify "prompt" or "messages"

What browsers are you seeing the problem on?
Latest Chrome and Firefox

Relevant log output from container

2025-06-17 13:27:28 error: [api/server/controllers/agents/client.js #sendCompletion] Operation aborted 400 Input required: specify "prompt" or "messages"
2025-06-17 13:27:28 error: [api/server/controllers/agents/client.js #sendCompletion] Unhandled error type 400 Input required: specify "prompt" or "messages"

2025-06-17 12:56:57 error: [api/server/controllers/agents/client.js #sendCompletion] Operation aborted 400 Provider returned error
2025-06-17 12:56:57 error: [api/server/controllers/agents/client.js #sendCompletion] Unhandled error type 400 Provider returned error

POST of "hostname/api/convos/gen_title" fails with 404 Not Found with the following reponse:
{"message":"Title not found or method not implemented for the conversation's endpoint"}

Screenshots

Let me know if we can workaround this error or to tweak the helm chart somehow.

[danny-avila]

See related comment here: #7818 (comment)

You are running into a model context window issue, and can be remedied via Agents/parameters

[iliyanpetrovcleverpine]

Thanks for your fast reply!

If I get it right I should create a new chat with the qwen/qwen3-32b and MCP selected, open the menu on the right, select Parameters, enter a number in the "Max Context Tokens" text box, save it as preset. Unfortunately after doing this I still get the same error. I tried values starting from 100 to 32768 but with no luck.
Could you help me understand if I am doing something wrong?

[iliyanpetrovcleverpine]

Providing "maxContextTokens" through the helm chart with "modelSpecs" did the job. Thanks again!

Edit: Actually I see that the problem is fixed if I provide "promptPrefix". Haven't tested if omitting the "maxContextTokens" but providing "promptPrefix" will work.