@tanstack/ai-anthropic weekly downloads statistics

@tanstack/ai-anthropic

Anthropic Claude adapter for TanStack AI chat, tool calling, thinking, and structured outputs

Installation

Setup

Set your API key as an environment variable:

Usage

Text/Chat Adapter

With Explicit API Key

With Tools

Thinking

Set how hard Claude thinks with reasoning on chat(). The adapter sends the right thinking fields for the model:

Models that think with a token budget, such as Claude Haiku 4.5, also take reasoning: { level: 'high', budgetTokens: 8000 }. The adapter raises max_tokens when it is below the budget. The types take only the levels and options of the selected model. See the docs for the levels of each model.

Prompt Caching

Mark a content part with cache_control to cache it:

Summarization

Provider Tools

Anthropic's native tools live on @tanstack/ai-anthropic/tools and go into chat({ tools }) alongside your own: webSearchTool, webFetchTool, codeExecutionTool, computerUseTool, bashTool, textEditorTool, memoryTool, and customTool.

Claude on Vertex AI

Install @anthropic-ai/vertex-sdk next to this package and import from @tanstack/ai-anthropic/vertex:

location is required (on the factory or via GOOGLE_CLOUD_LOCATION, GOOGLE_VERTEX_LOCATION, or CLOUD_ML_REGION); project can come from Application Default Credentials. Only the Claude models in the Vertex catalog are accepted.

Features

• ✅ Streaming chat • ✅ Function/tool calling and Anthropic's native provider tools • ✅ Extended and adaptive thinking • ✅ Structured output • ✅ Prompt caching • ✅ Summarization (anthropicSummarize) • ✅ Claude on Vertex AI (@tanstack/ai-anthropic/vertex)

Documentation

• Anthropic adapter: configuration, per-model thinking rules, every provider tool, and the API reference • Provider Tools: the concept, comparison matrix, and which models support which tools

License

MIT

Description from the npm listing.