Anthropic Claude Adapter

The Anthropic adapter provides direct integration with Anthropic’s powerful claude models via their REST Messages API, natively supporting tools, strict schemas, and dynamic conversational memory.

Unlike AWS Bedrock, this directly connects to api.anthropic.com using native API keys.

Configuration

You can enable Anthropic in your block’s config.json file:

{
  "adapter": {
    "provider": "anthropic",
    "model": "claude-3-5-sonnet-20241022",
    "maxTokens": 4096
  }
}

Supported Models

Any model compatible with Anthropic’s Messages API is supported, for instance:

• claude-3-5-sonnet-20241022
• claude-3-5-haiku-20241022
• claude-3-opus-20240229

Authentication

When configuring the block or running mblk init, Memoryblock will securely prompt for your active ANTHROPIC_API_KEY.

It will be saved in your ~/.memoryblock/auth.json (or your workspace’s local auth.json):

{
  "anthropic": {
    "apiKey": "sk-ant-api..."
  }
}

If you prefer using environmental variables for CI/CD or dockerized deployments, simply inject ANTHROPIC_API_KEY.

Sub-Agent Spawning

Because Memoryblock’s orchestration relies on create_agent, you can natively spin up Anthropic-backed specialized worker nodes mid-conversation by passing claude- prefixes in your tool payload:

User: Spawn a new agent called "data-cleaner" using Claude 3.5.

The system will automatically map it to the anthropic provider.