AI coding agents have quietly become part of most engineering teams. They open files, run commands, call tools, and push changes, often for hours at a time. Yet when you want to know what an agent actually did, the answer usually lives in one developer’s terminal history, and it disappears when that session closes.
Today we’re opening the beta of Scout Trails, a shared view of what your AI coding agents are doing across your team: every session, every tool call, and the cost behind it. The beta is free, and you can sign up here.
Why we built it
We’ve spent the last year connecting AI coding agents to production data through the Scout MCP server, so an agent fixing a slow endpoint or an N+1 query can see real traces and performance history instead of guessing from code. Along the way, the teams using it kept asking the reverse question: if the agent can see what happened in production, how do we see what happened inside the agent?
Individual developers can scroll back through their own sessions. Engineering leads have a harder time answering basic questions about how agents are being used across the team:
- Which developers are using agents, and how often?
- What did the agent actually do in the session that broke staging?
- Which tools are agents calling, and which workflows are sticking?
- Where is the token spend going, and is it growing faster than expected?
Trails is our answer to those questions.
What Trails shows you
Sessions across your team. Every agent session lands in one place, filterable by developer, so you can see adoption without asking anyone to export their history.
Tool calls inside each session. Drill into a session to see the sequence of tools and actions the agent took. When something behaves unexpectedly, you can see what happened instead of reconstructing it afterward.
Usage and cost. Token usage and dollar cost are tracked per session, so spend is visible as agent adoption grows rather than showing up as a surprise on the invoice.
Patterns over time. Because the data is shared and persistent, you can see which agents, tools, and workflows are becoming a real part of how your team builds software.
Setup is configuration only
Trails works with Claude Code today, along with any other coding agent that can export OpenTelemetry. There is no binary to download and no wrapper process to run. You set five environment variables to point your agent’s telemetry at Scout’s hosted ingest endpoint, and sessions start showing up.
Because it’s standard OpenTelemetry, Trails fits alongside whatever you already run rather than replacing it.
How Trails fits with Scout Monitoring
Trails is a separate product with its own signup, so you don’t need a Scout Monitoring account to use it. If you do use Scout Monitoring, the two cover both sides of AI-assisted development:
- Scout Monitoring and the Scout MCP server show what your code did in production, and make that context available to your agent while it works.
- Scout Trails shows what the agent did in your codebase to get there.
Together, you can move from “what did the agent change?” to “what did that change do in production?” without stitching the two together yourself. Connecting those two views more tightly is where we’re headed next.
Join the beta
Trails is free during the beta. We’re still shaping the product with early users, so we’d love to hear what you want to see next.
Get started with Scout Trails, read more on the Trails page, or reach us anytime at support@scoutapm.com.
