Microsoft Foundry Plugin for OpenClaw

Microsoft Foundry plugin

OpenClaw gains support for Microsoft Foundry as a model provider through this plugin.

Distribution

Package: @openclaw/microsoft-foundry
Installation method: bundled with OpenClaw

Surface

providers: microsoft-foundry; contracts: imageGenerationProviders

Image-generation provider: microsoft-foundry

Requirements

A Microsoft Foundry or Azure AI Foundry resource that has deployments set up.
Authentication via API key using AZURE_OPENAI_API_KEY or a provider API key that has been configured.
For Entra ID authentication, the Azure CLI must be installed and az login run before starting. OpenClaw keeps Microsoft Foundry runtime tokens current through az account get-access-token.

Chat models

Chat deployments in Microsoft Foundry rely on the provider model reference microsoft-foundry/<deployment-name>. During onboarding, the Azure CLI is used to locate Foundry resources and deployments, and the deployment name you pick gets written into the model configuration.

For supported OpenAI-compatible chat APIs, OpenClaw talks to the Foundry /openai/v1 endpoint:

GPT, o*, computer-use-preview, and DeepSeek-V4 model families use openai-responses by default.
MAI-DS-R1 and other chat-completion deployments fall back to openai-completions unless a specific supported API is set.
MAI-DS-R1 is flagged as reasoning-capable based on reasoning content rather than reasoning_effort. Its context and output token metadata sit at 163,840 tokens.

Anthropic Claude deployments within Microsoft Foundry follow the Anthropic Messages API format, not the OpenAI-compatible /openai/v1 one. Until the Microsoft Foundry plugin includes a native Anthropic runtime, configure these as a custom anthropic-messages provider. If the Foundry deployment name and the Claude model ID differ, set params.canonicalModelId on the model entry. That lets OpenClaw apply the right wire contracts per model, map /think off correctly, and keep signed thinking intact.

MAI image generation

The plugin registers microsoft-foundry for image_generate with the current Microsoft AI image models:

MAI-Image-2.5-Flash
MAI-Image-2.5
MAI-Image-2e
MAI-Image-2

Use a deployed MAI image deployment name as the model ref. Since the MAI API expects your deployment name in the request's model field, the provider does not set a default image model:

{
  agents: {
    defaults: {
      mediaModels: {
        image: {
          primary: "microsoft-foundry/<deployment-name>",
          timeoutMs: 600000,
        },
      },
    },
  },
}

Prompt-only generation hits Microsoft Foundry's MAI generations endpoint: /mai/v1/images/generations. For reference-image edits, /mai/v1/images/edits is called, and only MAI-Image-2.5-Flash and MAI-Image-2.5 deployments are supported.

When only the Foundry endpoint is configured, prompt-only generation can work with a custom deployment name. For image edits using a custom deployment name, pick the deployment via onboarding or supply model metadata so OpenClaw can confirm the deployment is powered by MAI-Image-2.5-Flash or MAI-Image-2.5.

Constraints for MAI images:

Output: a single PNG image per request.
Size: 1024x1024 is the default; width and height each need to be at least 768 px.
Total pixels: width × height cannot exceed 1,048,576.
Edits: one input image in PNG or JPEG format.
Shared hints that are unsupported, including aspectRatio, resolution, quality, background, and non-PNG outputFormat, are not forwarded to Microsoft Foundry.

Troubleshooting

az: command not found: get the Azure CLI installed or switch to API-key auth.
Microsoft Foundry endpoint missing for MAI image generation: choose a Foundry deployment during onboarding or add models.providers.microsoft-foundry.baseUrl.
supports MAI image deployments only: the chosen image model targets a deployment that is not MAI. For image_generate, use a deployed MAI image model.