How to analyze usage from your MCP Server
Two weeks ago Anthropic AI launched Model Context Protocol (MCP), an open source framework that defines a standard server-client architecture to make LLM applications context-aware.
We were one of the first adopters of MCP, and we built and tested our first server MCP Server just two days after the announcement from Anthropic. We think it can create a lot of opportunities to use Tinybird in new ways.
But vision was getting ahead of practice, and we thought it would be a good idea to put some basic observability infrastructure in place to monitor how our MCP server is being used. Something that would help us discover errors, understand how the tools were being used, and get a feel for how we could make v1 better than v0.
For now, MCP Servers are installed locally. Anthropic has mentioned they'll add SDKs for remote production servers in the future, but for now, it's local. Thus, there's nothing pre-built for remote logging collection and analysis.
So, after we built our MCP Server and started testing it locally (and encouraging our community to do the same), we quickly added logging and analysis features for metrics like:
