TrackMCP launched on September 4, 2026, giving anyone who ships a Model Context Protocol server a way to see who is actually using it. One line of SDK code captures client connections, tool calls, errors and whether a workflow finished, and the free tier costs nothing.
What Happened
Built by Krishna Goyal, TrackMCP wraps an existing server rather than replacing it. In TypeScript the integration is npm i @trackmcp/sdk followed by wrapping the server export in withTrackMCP with an API key and a service name. It supports the official TypeScript and Python MCP SDKs, and telemetry is captured at the server boundary, so it works with Claude, Cursor, ChatGPT and custom agent clients without per-client instrumentation.
What This Enables
Most people publishing a Model Context Protocol server today are flying blind. They know the server is up and nothing else. TrackMCP answers the four questions that decide what you build next: which clients connect and whether they come back, which tools get called and which are ignored, where sessions error out or retry, and whether the job the agent started actually completed. Add the wrapper, ship, and after a week you have a list of tools nobody touched and a list of schemas that agents keep getting wrong.
Why It Matters for Creators
MCP is how a creator tool gets reachable by an agent, and the number of servers is growing faster than anyone's ability to tell which ones work. If you have exposed a render pipeline, an asset library or a publishing workflow over MCP, completion rate is the metric that matters, not uptime. A tool that an agent calls and then abandons halfway is worse than one that is missing, because the agent burns tokens finding that out. TrackMCP also runs an MCP repository where servers can be browsed by category and by GitHub stars.
Key Details
Hobby: free, 1,000 captured tool calls per month, 7-day retention, 1 server, 1 team member.
Pro: 49 dollars per month, 50,000 tool calls, 90-day retention, up to 5 servers and 5 team members.
Enterprise: custom volume, retention and multi-team environments, priced on request.
SDKs: TypeScript and Python.
Captured: active clients, new connections, tool adoption, workflow completion, sessions, errors, retries, schema mismatches, retention.
What to Do Next
If you already run a server, install the SDK on a staging instance and leave it for a week before changing anything. The first useful output is the gap between tools you thought were the product and tools agents actually call. Our write-up of FetchSandbox for testing agent API integrations pairs with this: one tells you what broke in production, the other catches it before you ship.
