Skip to main content
Sedata.AI for MCP
@sedata-ai/mcp is a drop-in instrumentation package for Model Context Protocol servers. It captures every tool call as an OpenTelemetry span, ships it to a backend you control, and runs an inline safety check on tool inputs — all in one wrapper.

​
What you get

Zero-config tracing

Every registerTool call is auto-wrapped with OpenTelemetry spans, durations, and counters.

Inline safety checks

safetyCheck() blocks malicious or sensitive content before your handler runs.

OTLP-native

Ship to Sedata Cloud, your own collector, or any OTLP/HTTP backend.

Custom instrumentation

Drop into the underlying tracer and meter for your own spans, histograms, and counters.

​
How it fits

1

You wrap your MCP server

One call to instrumentServer(server, config) patches registerTool so every tool invocation runs inside an OTel span.
2

Tool calls produce spans + metrics

Each call records duration, status, request id, session id, client address, tool metadata, and (optionally) safety-check verdicts.
3

Data ships to your backend

OTLP/HTTP exporters batch traces and metrics out to the endpoint you configure. Use Sedata Cloud, an OTel collector, or a console exporter for local dev.

​
Two pieces, one package

instrumentServer()

The single entry point. Returns an ObservabilityInstance with helpers for custom spans, histograms, counters, and graceful shutdown.

safetyCheck()

A handler wrapper that runs your tool input through Sedata’s safety API before the handler is called.

Quickstart

Get a server emitting traces in five minutes.

API reference

Every function, class, and type exported from the package.

Examples

Working servers you can run end-to-end.
The Python and Go packages mirror the TypeScript SDK. This documentation covers the TypeScript surface; tracking issues for the other runtimes live on GitHub.