Key takeaways
- Clients differ in discovery, argument shape, and retry behavior.
- Break every core metric down by client.
- Growth concentrated in one client is a risk worth knowing.
Not all MCP clients behave the same. They differ in how they discover tools, how they format arguments, and how persistent they are on retries. Your client mix is a design input, not a vanity metric.
Why the mix matters
- A tool that works in one client may fail in another over argument shape
- Growth concentrated in one client is a risk worth knowing
- Custom agents often reveal power-user workflows worth supporting
What to do with it
Break every core metric down by client. If completion rate is high in Claude and low in Cursor, the problem is client-specific and so is the fix. If one client drives most growth, test changes against it first.
Client mix turns 'our server is doing fine' into 'our server is doing fine for these clients and not those,' which is the version you can act on.
See this on your own server
TrackMCP turns your MCP server's calls into adoption, workflows, and outcomes. One line to install.