October was about making Scout’s data easier for AI tools to use and easier for your team to trust. Scout is now in Claude’s connector directory, the MCP server got leaner and more careful about what it returns, and Scout Trails opened its beta. Anomaly Detection also became easier to read and better at catching short spikes.
Here’s what shipped in October:
Scout Is in Claude’s Connector Directory
You can now connect Scout to Claude on the web or the desktop app straight from Claude’s connector directory. Open the listing, click Connect, and sign in with your Scout account. Claude can then answer questions like “which endpoints got slower this week?” by pulling response times and traces from Scout, down to the SQL behind a slow request. The connector is read-only.
We wrote up what the listing means and what you can ask in Scout Monitoring Is Now in Claude’s Connector Directory. The hosted server is also listed in the official MCP Registry now, alongside the local server, so MCP clients and directories that read from the registry can find it.
Scout MCP: Leaner Responses, Paging, and Safer Defaults
Compact traces. GetAppTrace now returns a compact trace built for LLM context windows. Identical sibling spans collapse into one node with a count and average duration, redundant fields are dropped, and the tree is capped at the 100 slowest spans, with a note on how many were omitted. The REST API offers the same format with ?compact=true.
Breakdowns for job traces. Background job traces used to come back with an empty span list. They now include a per-layer timing and allocation breakdown and the slowest operations, with their SQL. Job trace ids also resolve through GET /api/v0/apps/:id/traces/:trace_id.
Paging. GetAppJobs, GetAppEndpointTraces, GetAppJobTraces, and GetAppAnomalyEvents now take limit and offset, report total_count and has_more, and default to 20 results.
Safer by default. ListOrgs no longer returns your Scout Key, so the agent ingest credential stays out of assistant conversations. Tool errors are flagged as errors with a clear not_found for unknown endpoint and job ids, internal errors no longer include backtraces, and timestamps are always UTC. Every tool also declares itself non-destructive, which some MCP clients use to decide what an agent may call without asking.
Released in: hosted server (automatic), scout-mcp-local 2026.10.20
Scout Trails Opens Its Beta
While Scout MCP gives your agent visibility into production, Scout Trails covers the other direction: what your AI coding agents are doing across your team. Every session, the tools the agent called inside it, and the token usage and cost behind it land in one shared view.
Trails works with Claude Code and any other coding agent that exports OpenTelemetry. Setup is five environment variables, with no binary to install. It’s a separate product with its own signup, and it’s free during the beta. The details are in Scout Trails Is in Beta.
Anomaly Detection: Severity at a Glance
Anomaly markers on endpoint and background job charts now show how far each one deviated, with chips like “3.8x” placed so they don’t overlap. Marker opacity scales with severity, so the anomalies that matter stand out on long timeframes. While you drag the sensitivity slider, a small indicator shows the threshold you’re setting.
Detection itself got more thorough. Scans now score every minute since the last scan, up to an hour back, instead of only the most recent minute. A short spike that landed between scans is now caught, and if it’s reported more than 10 minutes after it started, the notification says it was detected late so you know why it arrived when it did.
Agents and Platform
- Ruby logging now writes log text that isn’t valid UTF-8 instead of raising, so a request containing a stray byte can’t turn into a 500 through the logger. It also handles log records from newer OpenTelemetry Ruby SDK versions. Released in: scout_apm_logging 2.2.0
- Python logging now installs alongside Python agent 4.x. Released in: scout-apm-logging 1.0.4
- Elixir gains a
core_agent_ca_certoption, matching the Python agent 4.0 option, so the Core Agent can trust a private CA, such as one presented by a corporate TLS proxy. Released in: Elixir agent 2.2.0 - Node.js agent 2.2.1 updates its dependencies.
- DevTrace no longer leaks variables into your page’s global scope. A customer report showed the widget overwriting their app’s global
t()function, and it also no longer conflicts with RequireJS or Materialize. - Sign-in now handles very long return URLs, which could previously make login fail.
From the Scout Community
FastAPI Shipped Three Breaking Changes in 21 Hours: What to Check
FastAPI 0.146 through 0.148 deprecated root_path in the app constructor, started treating omitted boolean form fields as False, and stopped validating default values. None of them crash on startup, which is what makes them easy to miss. We tested every example against FastAPI 0.145 and 0.148 and listed the searches to run before upgrading.
GoodJob 4.22 Deprecates Concurrency Keys: How to Move to Labels
GoodJob 4.22 deprecates concurrency keys and max_threads ahead of v5. The migration to labels is mostly mechanical, except that jobs already in the queue when you deploy don’t carry the new label. It also covers why throttled jobs show up as errors in your APM, and what a steady stream of them tells you.
Questions or feedback about anything here? Reach us at support@scoutapm.com.