mcpspan

mcpspan is analytics for the MCP servers you build. Your server reports each call through an SDK in one of eight languages, or through a proxy with no code change, and mcpspan shows which tools agents call, how often they fail and why, how fast they answer, and which clients call them. Argument values never leave your server. This connector lets Claude answer from those numbers, so you can ask in plain language instead of digging through a dashboard: - Which tools failed most this week, and why? - How does release 1.4.1 compare with 1.4.0? - What do agents ask for that my server doesn't have, and what did they do next? - Does Cursor fail more often than Claude Code? - Which tools are agents shown but never call, and what do their definitions cost in context? - What happened in the session where booking failed? Every tool only reads. The connector sees what you see in your mcpspan dashboard and cannot change anything: not servers, keys, alerts or settings. You sign in with your mcpspan account and choose to allow Claude; Settings → Connected apps in mcpspan lists it and disconnects it at once. Works with mcpspan Cloud. mcpspan is open source (MIT) and can also be self-hosted.