How to Use OpenClaw with Anthropic API — Configure Claude

OpenClaw uses Anthropic's Claude models by default, and the integration goes deeper than just sending messages. You can configure model routing, enable extended thinking for complex tasks, tune token limits, and use Claude's advanced features like tool use and streaming — all through your OpenClaw config.

Basic Anthropic API Setup

# In ~/.openclaw/.env:
ANTHROPIC_API_KEY=sk-ant-api03-your-key-here

# In openclaw.json:
{
  "llm": {
    "provider": "anthropic",
    "model": "claude-opus-4-6",
    "maxTokens": 4096
  }
}

Available Claude Models

Model Routing by Task Type

# In openclaw.json:
{
  "llm": {
    "provider": "anthropic",
    "default": "claude-sonnet-4-6",
    "routing": {
      "planning": "claude-opus-4-6",
      "coding": "claude-opus-4-6",
      "classification": "claude-haiku-3",
      "summarization": "claude-haiku-3",
      "search-synthesis": "claude-sonnet-4-6"
    }
  }
}

Extended Thinking

Enable Claude's extended thinking for complex reasoning tasks:

# In your agent config or session spawn:
{
  "thinking": {
    "type": "enabled",
    "budget_tokens": 10000
  }
}

# Best use cases:
# - Complex code architecture decisions
# - Multi-variable business strategy
# - Debugging hard-to-trace issues
# - Research synthesis across many sources

# Note: uses more tokens — use selectively

Tool Use Configuration

Claude's native tool use powers OpenClaw's skill system:

# Tools are defined as JSON schema:
{
  "name": "web_search",
  "description": "Search the web for current information",
  "input_schema": {
    "type": "object",
    "properties": {
      "query": {"type": "string"},
      "count": {"type": "number"}
    },
    "required": ["query"]
  }
}

Token Optimization

# Reduce prompt token waste:
# 1. Keep SOUL.md concise — injected every request
# 2. Summarize MEMORY.md regularly — do not let it grow unbounded
# 3. Use system prompt caching for large prompts
# 4. For long conversations, summarize context after 10+ turns

# Monitor usage:
openclaw stats --tokens --last 7d

Common API Errors

# 401 Unauthorized -> check API key in ~/.openclaw/.env
# 429 Too Many Requests -> rate limit hit
# 529 Overloaded -> Anthropic backend busy, retry with backoff
# 400 Bad Request -> likely invalid model name or malformed request

The OpenClaw Playbook ($9.99) covers the full Anthropic integration including prompt caching setup, cost tracking dashboards, and model routing configs for common use cases.