andreisirbu91-lab / MCPSpend

What is MCPSpend?

MCPSpend is a real-time cost tracking layer for the Model Context Protocol. It wraps your existing MCP servers and streams every tool call to a hosted dashboard, breaking down spend by tool, project, and customer. The setup is one command: `npx @mcpspend/proxy init --key `, and it auto-detects Claude Desktop, Cursor, Windsurf, and VS Code. Under the hood it uses Express, Prisma, Postgres, and a Redis-backed queue to handle high-volume ingestion. Call attribution, latency tracking, and cost estimation happen automatically.

Why is it gaining traction?

The MCP ecosystem is exploding, and people are starting to ask "wait, how much is my AI actually spending?" MCPSpend answers that before you even think to look. The real-time SSE ticker on the dashboard is the hook—you watch dollars accumulate live. The cost forecasting with day-of-week seasonality actually works (it's not naive linear extrapolation), and the anomaly detection fires Slack alerts when a tool goes rogue at 2am. Budget alerts at 50/80/100% of your cap catch problems before they become invoice shocks. EU hosting matters for GDPR-sensitive shops.

Who should use this?

SaaS teams building on MCP who bill customers per tool usage will want the per-customer labeling. Agencies managing multiple client workspaces need the project scoping. Any developer running Cursor or Windsurf with multiple MCP servers should know their actual cost per session. Teams on the Pro or Team plan get audit logs, CSV export, and role-based access—critical for compliance.

Verdict

MCPSpend solves a real problem in an emerging space, and the implementation is more polished than a 1.0% credibility score suggests—clean architecture, thoughtful security, and actual product-market fit. That said, 18 stars means you are an early adopter. The free tier is generous enough to kick the tires, but treat self-hosting as experimental until the SOC 2 audit lands Q4 2026. Worth evaluating now; treat production usage as a calculated risk.