@f3d1/llmkit-mcp-server

Query your AI API costs from Claude Code or Cursor. See spend by model, session, or time range without leaving your editor.

Part of LLMKit, an open-source API gateway that tracks costs and enforces budgets across 11 LLM providers.

Setup

1. Create a free account at dashboard-two-zeta-54.vercel.app

2. Create an API key in the Keys tab

3. Add the server to your Claude Code or Cursor config:

{
  "mcpServers": {
    "llmkit": {
      "command": "npx",
      "args": ["@f3d1/llmkit-mcp-server"],
      "env": {
        "LLMKIT_API_KEY": "llmk_your_key_here"
      }
    }
  }
}

Paste into .mcp.json (project root) for Claude Code, or .cursor/mcp.json for Cursor.

Tools

llmkit_usage_stats : Total spend, request count, top models, cache hit rate for a time period (today/week/month).

llmkit_cost_query : Cost breakdown grouped by provider, model, session, or day. Supports filtering by provider or model.

llmkit_budget_status : Current budget usage and remaining balance across all budgets or a specific one.

llmkit_session_summary : List recent agent sessions with request counts, costs, duration, and models used.

llmkit_list_keys : Show all API keys with status and creation date.

llmkit_health : Ping the LLMKit proxy to check if it's reachable.

Environment variables

Example

Ask Claude Code: "How much have I spent on AI this week?"

LLMKit Usage (week)
---
Requests: 47
Total spend: $0.34
Input tokens: 128,400
Output tokens: 24,100
Cache hit rate: 42.1%

Top models:
  gpt-4o-mini: 31 requests
  claude-sonnet-4-20250514: 12 requests
  deepseek-chat: 4 requests

License

MIT