What problem does it solve? Teams running MCP servers across Claude Desktop, Cursor, Windsurf, and VS Code have no visibility into how much each tool call costs, making it impossible to attribute AI agent spending to specific tools, projects, or customers. ## Core Features & Use Cases - Transparent Proxy Wrapping: Automatically wraps any MCP server (stdio or HTTP/SSE) to record per-call cost, latency, and token metadata without changing server behavior. - Usage Query MCP Server: Exposes tools like get_today_cost, list_top_tools, get_tool_usage, and list_sessions so you can query spending directly from any MCP client. - Per-Customer Attribution & Budget Alerts: Attach metadata such as customer and project to each server, then configure thresholds and Slack/email alerts in the dashboard. - Use Case: An agency running MCP servers for multiple clients wraps each server with customer metadata, then queries monthly spend per customer to generate accurate invoices. ## Quick Start Ask the AI to set up MCPSpend cost tracking by running the init command with your API key to wrap all configured MCP servers.