Loading page
Loading page
Monitoring
See which tools AI agents call on your MCP server, how fast they answer, why they fail, and what the agent did next.
mcpspan is open-source analytics for MCP servers. MCP servers give AI agents tools to call. mcpspan records each call: which tool, which client, how long it took, how big the answer was, and whether it failed. A dashboard shows it all.
It looks at agent behaviour, not just server health. After a failure, it shows whether the agent retried, tried something else, repeated itself or gave up. It also counts the tokens your tool list costs at the start of each session.
mcpspan is for developers who build and run MCP servers, whether public ones or internal tools for their own agents. It fits teams who want to know if agents use their tools as planned.
Tool calls can carry private data from your users. A self-hosted mcpspan keeps everything in a TimescaleDB database in your own service, and nothing is sent anywhere you did not set up.
The project is young and moves fast, so read release notes before you upgrade. Keep API_KEY_SECRET private. You look after updates and backups of the database yourself.
Containers: timescale/timescaledb:2.30.1-pg17, ghcr.io/mcpspan/core-api:0.15.0, ghcr.io/mcpspan/core-dashboard:0.15.0
Step 1
Choose the Starter plan. It costs $20 a month and runs up to 3 apps.
Step 2
Open App templates in the dashboard, choose mcpspan, and click Deploy. We set up the containers, secrets and domain.
Step 3
Wait a few minutes for the first deploy, then open your new URL and finish the mcpspan setup screen.