Application performance monitoring company Sentry Inc. today announced the launch of MCP Server Monitoring, a new service that gives anyone building on top of the Model Context Protocol a clearer view into what is working and not working behind the scenes.

MCP, introduced by Anthropic PBC last November, has seen rapid adoption as companies move to make their products artificial intelligence-compatible and turn to MCP to give AI agents a safer and standardized way to use their tools and data. Despite strong uptake, Sentry argues, though, that once an MCP service is live, developers generally have limited visibility into how it’s working.

“We built this because we needed to debug problems in our own MCP server and quickly learned they’re the same problems everyone building MCPs is having,” explained Cody De Arkland, head of developer experience at Sentry.

“Existing monitoring tools struggle with the context of what’s happening in an MCP server,” De Arkland added. “We needed to know things like traffic load and AI client usage, which tools were getting called the most, which were slow or failing, and which inputs were causing things to break. We needed to know all of this without relying on users to tell us.”

With just a few lines of code, Sentry’s MCP monitoring helps development teams quickly answer key questions, such as which AI clients (such as Cursor or Claude) are sending requests, which tools are getting the most use and which ones are running the slowest or producing errors. Additionally, development teams can also find out why error rates are suddenly increasing and whether they coincide with traffic spikes or new releases, if problems are caused by recent changes or by bots sending malformed requests and whether errors are isolated to a specific transport, such as HTTP clients timing out while stdio remains unaffected.

Sentry brings its own scale-tested experience to the table, having run its MCP server at more than 30 million requests per month across thousands of users. The volume has exposed gaps in existing monitoring tools, such as difficulty pinpointing performance bottlenecks in specific tool calls, tracing silent failures that appear fine at the infrastructure level and understanding exactly which clients and transports are affected by errors.

MCP Server Monitoring addresses these gaps by breaking down usage by transport type (HTTP, SSE, stdio), showing top-used and error-prone tools and allowing teams to drill down to individual JSON-RPC calls, including their arguments and results.

The service is also designed to surface problems before customers notice them. For example, it can detect when a new deployment causes a 60% drop in a specific method’s usage, reveal that the issue is limited to a single transport, or flag an unknown client flooding a server with 100% failing requests. It also helps teams prioritize development by showing exactly which features and tools are used most, avoiding wasted effort on underutilized capabilities.

Today’s launch is the just the beginning from MCP Server Monitoring, with Sentry planning to expand the service with features such as Trace Propagation to improve downstream performance visibility, broader platform integrations including Cloudflare Inc.’s McpAgent and support for additional languages like Python.

Image: SiliconANGLE/Reve