Clawback MCP Server

LLM cost optimization tools for AI agents via the Model Context Protocol. Audit your spend, activate smart routing, and track savings without changing your code.

Quick start (no install)

Add to your MCP client config:

{
  "mcpServers": {
    "clawback": {
      "url": "https://clawback.run/mcp"
    }
  }
}

Local server

npx @clawback/mcp-server

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "clawback": {
      "command": "npx",
      "args": ["@clawback/mcp-server"]
    }
  }
}

Claude Code

Add to your Claude Code settings:

{
  "mcpServers": {
    "clawback": {
      "command": "npx",
      "args": ["@clawback/mcp-server"]
    }
  }
}

Tools

| Tool | Description | Auth | |------|-------------|------| | start_audit | Start a free LLM cost audit. Set one env var, Clawback observes your API calls and generates a savings report. | API key | | get_audit_status | Check audit progress, call count, report URL, and routing recommendations. | API key | | activate_routing | Activate smart routing after audit completes. Strategies: balanced, max_savings, conservative. | API key | | optimize_llm_call | Route a completion to the best price-performance model for the task type. | API key | | get_frontier | Get the efficient frontier (best quality per dollar) by task type. | None | | check_savings | Check monthly and all-time routing value. | API key | | get_model_recommendation | Get a model recommendation for a task type without making an API call. | None | | set_agent_budget | Set a monthly budget limit per agent with action on exceed (block, downgrade, alert). | API key |

How it works

1. Audit first. Call start_audit to register. Set your OPENAI_BASE_URL to https://clawback.run/v1. Your API calls flow through Clawback's proxy (zero latency impact, your keys, your data).

2. Get your report. At 50 calls, Clawback generates an audit report. It replays your calls against cheaper models, scores quality per task type, and tells you exactly how much you can save.

3. Activate routing. Review the report. If the savings are real, call activate_routing. Every subsequent call gets routed to the optimal model on the efficient frontier.

4. Track savings. Use check_savings to see your monthly routing value. Free until routing value exceeds $500/month, then 15% of the price difference per routed call.

Efficient frontier

Clawback maintains a real-time efficient frontier of LLM models, borrowed from modern portfolio theory. For each task type (extraction, classification, summarization, function calling, rewriting), it maps every model on a quality vs. cost chart and identifies the Pareto-optimal set. Your calls get routed to the frontier.

Use get_frontier to see the current frontier for any task type, no auth required.

Security

• API keys are SHA-256 hashed for customer identification and never stored in plaintext
• BYOK: your API keys, your data, Clawback never bills for LLM usage
• All traffic over TLS, keys pass through proxy but are never persisted
• 30-day response body retention, 90-day metadata retention

Pricing

• Free audit, always
• Smart routing: 15% of the price difference per routed call
• Free until routing value exceeds $500/month
• You keep 85% of every dollar saved

Links

• Website: clawback.run
• Efficient Frontier: clawback.run/frontier
• Security: clawback.run/security