Session timeline
Follow every MCP tool execution in order, including retries, failures, latency spikes, and downstream calls.
Observability for MCP servers.
Add one line to your MCP server and trace every tool call, input, output, error, and latency in real time.
Built for MCP today. Expanding into full AI runtime observability.
100%MCP tool calls captured in real time
<5 minTypical setup time
1 viewInputs, outputs, errors, latency, and replay
prod-readyBuilt for production MCP tools
See every MCP tool call in one timeline.
Scope captures MCP tool inputs, outputs, errors, latency, and execution flow in a single trace so you can understand exactly what happened during a run.
Tool call context
Inspect tool inputs, outputs, errors, and latency to understand exactly where execution failed.
Security & authorization
Audit transport channels, decode active authorization scopes, and verify authenticated user identities.
From first event to root cause.
Find the failed MCP call, inspect the exact input and output, identify the issue, and ship the fix quickly.
Capture
Wrap your MCP server with the Scope SDK.
Trace
View every MCP tool call with timing, status, and latency.
Inspect
Open any execution to inspect inputs, outputs, logs, and errors.
Resolve
Identify slow tools, schema mismatches, auth failures, and bad downstream responses.
An SDK that stays out of the way.
Instrument your MCP server in one line and automatically capture tool calls, inputs, outputs, latency, and failures.
Designed for production MCP servers
Instrument tool execution, retries, errors, latency, and custom context with zero impact on your tool logic.
JS / TS:
npm i @scope-gs/sdk
Python:
pip install scope-sdk
mcp-server.ts@scope-gs/sdk
import { init } from "@scope-gs/sdk"; init(server, { apiKey: process.env.SCOPE_API_KEY, });
Every MCP tool call is automatically traced.
The view your on-call needs at 2am.
Open the failed MCP call, inspect the exact input and output, and understand what happened immediately.
run_01JDX9 · search_docs authorization failureMissing API token rejected by downstream MCP tool
ErrorStatus
run_01JDX8 · crm.lookup_customer timeoutCustomer lookup request exceeded timeout threshold
SlowStatus
run_01JDX7 · slack.create_ticket completed successfullyChannel notification sent successfully
OKStatus
FAQ
Built for the messy reality of MCP production.
What does Scope capture?
Scope captures MCP tool calls, inputs, outputs, latency, errors, logs, retries, and execution metadata.
Does Scope work with our existing MCP server?
Yes. Add the SDK and Scope automatically traces tool execution without changing your tool logic.
Will Scope support full AI agent tracing?
Yes. Scope starts with MCP observability and expands into broader AI runtime tracing across LLM calls, workflows, and agents.
Start tracing your MCP server.
Install Scope, instrument your MCP tools, and understand every execution path in minutes.