Custom azure open AI hosted model + MCP server - azureOpenAIApiInstanceName is required when using azureOpenAIApiKey

[swainsa1]

Hello @danny-avila , I am trying Librechat out with our custom internal hosted LLM , I was able to successfully set that up using an azureOpenAI end points ,

• Custom base url using proxy - http://localhost/openai/deployments/gpt4o_2024-05-13 -> Which internally routes the request to our azure hosted instance
• version property set to "2024-10-21" - Which is sent as part of the query param
  -apiKey- is sent as the bearer token . Or I have the option of passing a http_api_key to my proxy where I can always attach as bearer token header

I also wanted to integrate an in built MCP server. I was able to hook that up also in the librechat.yaml configuration .

• I see the listing of all the tools when the server is starting (Running as a standard docker container)

Issue :

After the MCP server integration. when I start the nom backend:dev and ask any question I get an error
azureOpenAIApiInstanceName is required when using azureOpenAIApiKey
with debug enabled its pointing to a AzureChatOpenAI._getClientOptions ()

Any help on this would be greatly appreciated.

2025-05-09T21:06:40.130Z error: [api/server/controllers/agents/client.js #sendCompletion] Operation aborted azureOpenAIApiInstanceName is required when using azureOpenAIApiKey
2025-05-09T21:06:40.130Z error: [api/server/controllers/agents/client.js #sendCompletion] Unhandled error type azureOpenAIApiInstanceName is required when using azureOpenAIApiKey

Error: azureOpenAIApiInstanceName is required when using azureOpenAIApiKey
at Object.getEndpoint (XXX/LibreChat/node_modules/@librechat/agents/node_modules/@langchain/openai/dist/utils/azure.cjs:42:19)
at AzureChatOpenAI._getClientOptions (XXX/LibreChat/node_modules/@librechat/agents/dist/cjs/llm/openai/index.cjs:109:37)
at AzureChatOpenAI.completionWithRetry (XXX/LibreChat/node_modules/@librechat/agents/node_modules/@langchain/openai/dist/chat_models.cjs:2336:37)
at AzureChatOpenAI._streamResponseChunks XXX/node_modules/@librechat/agents/node_modules/@langchain/openai/dist/chat_models.cjs:1941:43)
at _streamResponseChunks.next ()
at AzureChatOpenAI._generateUncached (XXX/LibreChat/node_modules/@langchain/core/dist/language_models/chat_models.cjs:215:34)
at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
at async AzureChatOpenAI.invoke (XXX/LibreChat/node_modules/@langchain/core/dist/language_models/chat_models.cjs:92:24)
at async RunnableCallable.func (XXX/@librechat/agents/dist/cjs/graphs/Graph.cjs:382:39)
at async RunnableCallable.invoke (/LibreChat/node_modules/@langchain/langgraph/dist/utils.cjs:82:27) {
pregelTaskId: 'b6a8b7c9-06d9-54ae-8462-39f143556600'

LibreChat.yml
endpoints:
azureOpenAI:
# Endpoint-level configuration
titleModel: "gpt-4-turbo"
plugins: true
assistants: false
groups:
# Group-level configuration
- group: "XXXXX"
apiKey: "XXXX"
deploymentName: "gpt-4o_2024-05-13"
version: "2024-10-21"
baseURL: "http://localhost/openai/deployments/gpt-4o_2024-05-13/"

    serverless: true
    # Model-level configuration
    models:
      gpt-4: true

MCP Server configuration
mcpServers:
mcpServerName:
type: stdio
command: docker
args:
- run
- -i
- --rm
- -e
- TOKEN=${TOPKEN}
- our_mcp_image:latest
env:
TOKEN: "XXXXXX"

Based on my research looks like code is failing somewhere in this method

LibreChat/api/server/controllers/agents/client.js

Line 545 in 4af72aa

async chatCompletion({ payload, abortController = null }) {

[swainsa1]

This Issue resolved after the latest code change pulled from last week