Skip to main content

Scout MCP: Query Your Observability Data Through AI Assistants

· 3 min read
Nimisha G J
Nimisha G J
Engineer at base14

Scout supports the Model Context Protocol (MCP). You can connect your coding agent to Scout and query traces, logs, metrics, service topology, and alerts using natural language.

What This Means​

MCP is a standard that lets coding agents call external tools. Scout's MCP endpoint exposes 10 read-only tools that cover the core observability operations: listing services, mapping dependencies, querying traces and logs, discovering schemas, checking metrics, and pulling alerts.

The tools are designed to be composable. An coding agent can chain multiple tools in a single investigation, discovering available span attributes before querying traces, or correlating error logs with the traces that produced them.

How It Works​

You connect your coding agent to Scout's MCP endpoint once. After that, you ask questions in plain English and your coding agent determines which tools to call.

A few examples of what you can ask:

  • Show me the slowest traces for payment-service in the last hour
  • Find error logs containing "timeout" in order-service today
  • What services does checkout-service depend on?
  • Were there any critical alerts in the last 2 hours?

The assistant handles tool selection, parameter formatting, and result interpretation. You get back a summary instead of raw JSON.

Supported Agents​

Setup takes a few minutes for each client:

  • Claude Code CLI adds the MCP server with a single command and authenticates via browser
  • Claude Desktop requires a config file edit and app restart
  • Gemini CLI requires a config file edit, with OAuth handled automatically

Full setup instructions are in the MCP Client Setup documentation.

Multi-Step Investigations​

The interesting part is how your coding agent combines tools. When you ask "the checkout page is slow, can you investigate?", your coding agent does not just run one query. It lists services to find the relevant one, queries traces filtered by high duration, fetches full span details for the slowest trace, and identifies the bottleneck.

Similarly, after a deployment, you can ask "we deployed order-service 30 minutes ago, are there any new errors?" and your coding agent will check traces, logs, and alerts for that time window, then summarize the findings.

One pattern worth highlighting: Scout MCP includes discovery tools for both spans and logs. These return available attribute keys, span names, severity levels, and status codes for a given service. The coding agent can use these to build precise queries without you needing to know the exact attribute names.

You can ask "what span names and attributes are available for payment-service?" before asking "show me traces where provider is stripe."

Read-Only by Design​

All 10 tools are read-only and idempotent. The coding agent cannot create, modify, or delete anything in your environment. This keeps the integration safe for everyday use without requiring additional permission scoping.

See It in Action​

Getting Started​

The MCP Client Setup guide covers connecting Claude Code, Claude Desktop, and Gemini CLI. The Usage Guide covers what you can ask, example investigations, and a full tool reference.