Hi, Microsoft just made Anthropic models available in Azure AI Foundry. I would like to utilize these models within n8n AI Agents but we currently only have option for Azure OpenAI Chat Model. Is there a way to use Azure-hosted Claude models within n8n AI Agents or we have to wait for the integration?
Is it possible to use newly added sonnet models with Azure AI Foundry using Azure OpenAI Chat Model?
5 Likes
Seconding this request.
3 Likes
Third. Do we know if they are working on this?
2 Likes
Fourth. Anthropic Chat Model Node doesnt work either when your credentials came from Azure AI Foundry
1 Like
Fifth, really hope the team can work on this in the near future.
2 Likes
Sixth, would be amazing to support all Foundry endpoints, especially non-OpenAI model selection.
1 Like
Seventh, i really need this 
Really would appreciate this! Happy to support any integration effort
Eighth this, really need this for my work.
I agree, the Azure “OpenAI” Chat Model should support Foundry Models and Endpoints. Currently it is way to limited for enterprise customers.
Would really appreciate this being addressed!
Hi all, I got this working. Add an Anthropic credential and set the baseUrl like the screenshot below. Note the connection will fail, but it will work in your flow.
Select an Anthropic Chat Model for your AI agent node as opposed to an Azure Open AI node. Choose your cred and use the “By ID” setting in the dropdown. Type your model deployment name in the text box. Execute and it should work. I only tried with API key auth, but this is working for me.
2 Likes
Can confirm that this does work, just connected to my Opus 4.6 and Sonnet models in Azure AI Foundry. Thanks!!!
1 Like
I lied… after two weeks of using Azure anthropic claude 4.7 and sonnet 4.6 with this hack job connection, streaming broke out of nowhere and now it takes minutes for any responses. Streaming would give me a preamble in my front end, and then 2min later, data dump the response in the chatbot. Switching back to AOAI 5.4 and 5.4-mini, streaming was back and fast
So the workaround I found works best is that you build a LiteLLM and point the native OpenAI nodes to that:The n8n bug — Issue #28635 (April 17, 2026):
The @n8n/n8n-nodes-langchain.lmChatAnthropic node v1.3 (which is exactly what you’re using) emits the legacy thinking format thinking: { type: "enabled", budget_tokens: N }. Anthropic has removed that format on Claude Opus 4.7 the API returns a 400 the moment thinking is on. The same format is also deprecated and scheduled for removal on Opus 4.6 and Sonnet 4.6. Anthropic now requires thinking.type.adaptive and output_config.effort to control thinking behavior. GitHub + 2
This is in n8n’s Linear queue with the AI team assigned, not yet shipped.
There’s also a separate n8n streaming bug — Issue #23851:
Enabling the “Streaming” toggle on the AI Agent does not pass the stream: true parameter to the underlying Anthropic SDK. This causes requests with large input token counts (~20k+) to fail with the error: “Streaming is required for operations that may take longer than 10 minutes.” This one’s been open longer and is exactly your shape long input contexts (your CPA agent’s system prompt alone is huge, plus 20-turn Redis memory, plus your prep context block) flip the request into a path where stream-flag propagation breaks. GitHub
And Anthropic confirmed model-side behavior changes:
All users now default to xhigh effort for Opus 4.7, and high effort for all other models as of April 7. Opus 4.7 also has a notable behavioral quirk relative to its predecessor: as we wrote about at launch, it tends to be quite verbose. Combined with extended thinking running silently in the background, that explains why long answers feel like they pause forever and then dump Opus 4.7 is doing more thinking than 4.6 did at the same effort tier.
LiteLLM proxy in your AKS cluster. 30-minute setup. Stand up a LiteLLM pod, point it at Anthropic with your real key, configure it to expose an OpenAI-compatible endpoint. Then in n8n, use the OpenAI Chat Model node with a custom base URL pointing at your LiteLLM service. LiteLLM handles all the streaming translation and adaptive thinking parameter shapes correctly — it’s literally maintained for this exact problem. You bypass the broken n8n Anthropic node entirely. This gets you Opus 4.7 back with proper streaming.
config.yaml
model_list:
-
model_name: claude-opus-4-7
litellm_params:
model: anthropic/claude-opus-4-7
api_key: os.environ/ANTHROPIC_API_KEY -
model_name: claude-sonnet-4-6
litellm_params:
model: anthropic/claude-sonnet-4-6
api_key: os.environ/ANTHROPIC_API_KEY
general_settings:
master_key: sk-your-internal-litellm-key
2 Likes