Description

Adds documentation for the Dapr Agents AnthropicChatClient, which brings the client in line with the other LLM backends (dapr/dapr-agents#733). Until now the docs had no page for Anthropic.

• New page developing-ai/dapr-agents/integrations/providers/anthropic.md. It covers:
  • Installation: the Anthropic SDK ships with dapr-agents, so there is no extra to install.
  • Configuration: ANTHROPIC_API_KEY, ANTHROPIC_MODEL (default claude-sonnet-4-6), ANTHROPIC_BASE_URL, timeout, and the default max_tokens=4096.
  • Direct generate(...) calls, including passing extra options through to the API (for example thinking).
  • Using the client as the llm of a DurableAgent.
  • Streaming with stream=True and the on_chunk callback: which chunks are produced, their metadata, and how thinking data is collected.
  • Image input: OpenAI-style image_url blocks (HTTP(S) URLs and base64 data URIs) are translated to Anthropic image blocks. The page lists the supported media types.
  • Tool calling: AgentTool is formatted for Anthropic, tool_choice accepts either form, and the page notes a limit on dict tools.
  • Structured output: json and function_call modes, list[Model], the behaviour when combined with streaming, and a note on the mode a DurableAgent uses internally.
  • Prompty with type: anthropic.
• dapr-agents-core-concepts.md: adds AnthropicChatClient to the list of LLM clients and links to the new page.

All behaviour was checked against dapr_agents/llm/anthropic/ on dapr-agents main.

Issue reference

Documents dapr/dapr-agents#733.

Checklist

• Code rules and doc conventions followed (front matter and tabpane/alert shortcodes match the neighbouring Drasi provider page)
• ref links point to existing pages
• Built locally with Hugo

🤖 Generated with Claude Code