What is MCP observability?
MCP observability is monitoring and debugging built for the Model Context Protocol. Instead of watching the HTTP requests a general-purpose APM sees, it traces the units of work that actually flow between an AI client and an MCP server: tool calls, resource reads, prompt fetches, sessions, and which client made each call. When an agent calls a tool that fails, MCP observability shows the tool name, the arguments, the response or error, the latency, and the session it belonged to. That is the level of detail you need to operate an MCP server in production, and it is invisible to tools that only see the HTTP envelope.
What it covers
An MCP exchange is a JSON-RPC message, not a tree of spans. MCP observability captures those messages directly: every tool call with its arguments and result, every resource read, every prompt fetch, grouped into the session that ties a client's sequence of calls together. The dimensions that matter, tool name, client identity, and session, are first-class, so you can break error rate and latency down by each of them.
Why an APM is not enough
Datadog, Sentry, and New Relic are excellent at HTTP, infrastructure, and stack traces. But one HTTP request can carry an MCP tool call that failed at the protocol level while the HTTP layer reports a clean 200. The tool name and arguments are an opaque request body to an APM, and a JSON-RPC error inside a 200 OK counts as a success. The result is a real failure your monitoring calls healthy. We cover this gap in detail in MCP observability vs APM.
How it works
Spanly captures the JSON-RPC packets that flow between client and server, either through a drop-in SDK for TypeScript and Python or a CLI that wraps any MCP server with no code change. It is additive: keep sending HTTP and infrastructure telemetry to your existing APM, and send MCP-shaped traffic to Spanly. Because the SDK propagates the W3C trace context on inbound MCP requests, every Spanly view links straight back to the matching trace in the APM you already run.
Getting started
Adding MCP observability takes a few lines. See how to monitor your MCP server in production for the practical walkthrough, the docs for the full setup, or start free on the pricing page.