Setting Up OpenClaw with Anthropic API Key (Post-OAuth Ban)

Configuring OpenClaw with your Anthropic API key is the most common LLM setup in the community. Claude's instruction-following capability and long context window make it excellent for complex agent workflows. This guide covers getting your API key, choosing the right model, and configuring OpenClaw correctly for 2026.

Getting Your Anthropic API Key

1. Visit console.anthropic.com and create an account or log in
2. Navigate to API Keys in the left sidebar
3. Click Create Key, name it (e.g., "OpenClaw Personal")
4. Copy the key immediately — Anthropic shows it only once
5. Store it in a password manager before pasting it anywhere

Your key format looks like: sk-ant-api03-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Available Claude Models for OpenClaw

Model	Best For	Cost (per 1M input tokens)
claude-3-5-sonnet	Best overall, complex tasks	$3.00
claude-3-5-haiku	Fast responses, simple tasks	$0.80
claude-3-opus	Deep reasoning, research	$15.00

For most personal OpenClaw agents, claude-3-5-sonnet is the sweet spot of capability and cost.

Configuring OpenClaw

Edit ~/.openclaw/openclaw.json:

{
  "llm": {
    "provider": "anthropic",
    "api_key": "sk-ant-api03-YOUR_KEY_HERE",
    "model": "claude-3-5-sonnet-20261022",
    "max_tokens": 8192,
    "temperature": 0.7,
    "context_window": 200000
  }
}

Understanding Context Window Costs

Claude's 200K context window is one of its major advantages for agent workflows — your agent can "remember" much longer conversations and process large documents. However, the entire context window is billed on every API call. If your agent accumulates a 50K token conversation history, each new message costs 50K input tokens plus the tokens in the new message.

Monitor your context usage and configure OpenClaw's context truncation setting to prevent runaway costs:

"context_truncation": {
  "enabled": true,
  "max_tokens": 50000,
  "strategy": "summarize_oldest"
}

Rate Limits

Anthropic imposes rate limits based on your usage tier:

• Tier 1 (new accounts): 40K tokens/minute
• Tier 2 ($40+ spent): 80K tokens/minute
• Tier 3 ($200+ spent): 160K tokens/minute

For a single personal agent, Tier 1 is typically sufficient.

Testing Your Configuration

After saving the config, test the connection:

python -m openclaw test-llm

This sends a simple test message and prints the response, confirming your API key and model configuration are working.

Frequently Asked Questions

What happens if I exceed my Anthropic rate limit?

OpenClaw automatically retries rate-limited requests with exponential backoff. You'll see slightly delayed responses but no errors in normal usage.

Can I set a monthly spending cap with Anthropic?

Yes. In the Anthropic console under Billing, set a monthly usage limit. OpenClaw requests stop once the limit is hit, preventing unexpected bills.

Should I use claude-3-5-haiku for cost savings?

For simple tasks (answering questions, short writing), Haiku is a great choice. For complex multi-step agent workflows, Sonnet's better reasoning is worth the higher cost.