Know when your MCP server breaks
An MCP server can answer HTTP requests and still be useless. The token can expire, a tool can be renamed in a release, or a tool can start failing for every client. Your users find out when their AI assistant stops working. An MCP check in elmah.io Uptime Monitoring finds out first.
A normal uptime check requests a URL and looks at the status code. That tells you very little about an MCP server, where the endpoint can respond while every tool is broken.
An MCP check speaks the protocol. From each region you pick, elmah.io connects to your endpoint, verifies that the tool you chose still exists, and calls it without arguments. The check fails if the server is unreachable, rejects the credentials, no longer has the tool, or the tool returns an error.
When a check fails, the error from your server is included, so you can see whether the token was rejected, the tool is gone, or the tool itself blew up.
Create an uptime check and choose MCP as the type. Enter the URL of your MCP endpoint, like https://example.com/mcp, and add an Authorization header with the value Bearer your-token if your server requires one.
Click Connect to MCP Server and elmah.io lists the tools your server exposes. Pick the one to monitor, choose your regions and timeout, and save. Pick a tool that needs no input and has no side effects, like a status or ping tool, since the check calls it without arguments.
Read the documentation for the full walkthrough and troubleshooting.
MCP checks run on the same schedule and from the same regions as the rest of Uptime Monitoring. A failing check is logged as an error in your elmah.io log, so your existing email, Slack and Microsoft Teams notification rules just work.
Read more about Uptime Monitoring, or see how the rest of the AI features fit together.