MCP Openrouter troubleshooting

[sla-te]

What happened?

{"code":400,"message":"Provider returned error","metadata":{"provider_name":"Fireworks","raw":"{\"error\":{\"object\":\"error\",\"type\":\"invalid_request_error\",\"code\":\"invalid_request_error\",\"message\":\"JSON Schema not supported: could not understand the instance `{'not': {}}`.\"}}"}},"headers":{},"level":"error","message":"[api/server/controllers/agents/client.js #sendCompletion] Unhandled error type 400 Provider returned error","pregelTaskId":"be164570-e48d-5765-9728-3cca87809188","requestID":null,"retriesLeft":6,"stack":"Error: 400 Provider returned error\n    at APIError.generate (/app/node_modules/@langchain/openai/node_modules/openai/core/error.js:45:20)\n    at CustomOpenAIClient.makeStatusError (/app/node_modules/@langchain/openai/node_modules/openai/client.js:158:32)\n    at CustomOpenAIClient.makeRequest (/app/node_modules/@langchain/openai/node_modules/openai/client.js:301:30)\n    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)\n    at async /app/node_modules/@langchain/openai/dist/chat_models.cjs:2411:29\n    at async RetryOperation._fn (/app/node_modules/p-retry/index.js:50:12)","status":400,"timestamp":"2025-08-24T20:49:09.649Z"}

Version Information

ghcr.io/danny-avila/librechat-dev                latest    19f761bb34c3   37 hours ago    1.56GB
ghcr.io/danny-avila/librechat-rag-api-dev-lite   latest    2f812738600a   7 days ago      1.67GB

Steps to Reproduce

1. Create Openrouter account
2. Create preset for e.g. GPT-5-Nano
3. Integrate According to docs
4. Integrate any MCP (e.g. fetch from official mcp repo)
5. No joy

What browsers are you seeing the problem on?

Chrome

Relevant log output

https://github.com/user-attachments/assets/9bad2fb1-f140-4204-974e-a53a494b1ef2

Screenshots

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

"Provider returned error" is an Openrouter-side error coming from the provider they are proxying.

MCP servers will "work" depending on the provider/model.

[danny-avila]

gpt-5-nano as you mentioned in the issue you opened:

[sla-te]

Interesting.

I have e.g.

mcpServers:
  github:
    type: streamable-http
    url: "https://api.githubcopilot.com/mcp"
    headers:
      Authorization: Bearer github_pat_REDACTED
    initTimeout: 90000
    timeout: 60000
    startup: true

endpoints:
  custom:
    - name: "OpenRouter"
      apiKey: "${OPENROUTER_KEY}" # NOT OPENROUTER_API_KEY
      baseURL: "https://openrouter.ai/api/v1"
      models:
        default: ["@preset/gpt-5"]
        fetch: true
      titleConvo: true
      forcePrompt: false
      # As per docs I saw, that dropParams is not required anymore, tested with and without
      dropParams: ["stop"]
      modelDisplayLabel: "OpenRouter"
modelSpecs:
  enforce: false
  prioritize: true
  addedEndpoints:
    - OpenRouter
  list:
    - name: openrouter-gpt-5
      label: "GPT-5 (OpenRouter)"
      default: false
      description: "GPT-5 via OpenRouter - Latest OpenAI model"
      preset:
        endpoint: "OpenRouter"
        model: "@preset/gpt-5"
        
And am getting this error consistently

[sla-te]

And they do work fine if I connect to any endpoint directly, for pretty much all of them (I have openai, anthropic, google and also portkey)

[sla-te]

Oh, yeah I see. It's the presets (docs) (@preset/gpt-5) that fail completely apparently. I just tried with openai/gpt-5-mini and it works perfectly. It got as far as that I explicitly disabled fallback and it routed me to GPT-4 labelled as Deepseek instead of gpt-5-mini.

[sla-te]

🤣

[sla-te]

And thats with this preset: