MCP Anomaly Incidents

Added

• MCP get_incident tool now supports both exception and anomaly incidents

Fixed

• Ensure the graph export popover isn’t cut off when placed at the bottom of the page.

MCP and product updates

Added

• Added a link to the docs in the trace timeline overlay explaining how to add additional instrumentation
• Added Anomaly incidents MCP tool
• Added a toggle switch to show OpenTelemetry internal attributes

  

Changed

• We've changed the MCP Token UI to better reflect the allowed tools. Instead of sections that can have read/write/delete rights, we're now showing each individual tool with a checkbox for exposure.

  

Improve error reporting for Next.js app

Fixed

• Report errors in AppSignal for exception events in Next.js render spans.

This release can be installed through our collector packages and Docker image.

Log trigger filtering

Added

• Add functionality that allows filtering log triggers by name, source, or creator.

  

Support streaming bodies and other improvements

Added

• Add Sidekiq worker-level job status metric: worker_job_count. This new counter metric's status tag will be processed for each job that's processed and reports another counter with the failure status if the job encountered an error.

Fixed

• Support streaming bodies. AppSignal's Rack instrumentation now supports streaming bodies in responses, such as those produced by Async::Cable. This fixes an issue where AppSignal's Rack instrumentation would cause requests with streaming bodies to crash.

  If you use our Rack instrumentation through a framework that is automatically instrumented by AppSignal, such as Rails, Hanami, Padrino or Sinatra, this fix is applied automatically.

  If your application instruments Rack manually, you must remove the following line from your application's initial setup:

  Ruby

  use Rack::Events, [Appsignal::Rack::EventHandler.new]

  And replace it with the following line:

  Ruby

  use Appsignal::Rack::EventMiddleware

• Avoid instrumenting Rails when AppSignal is not active. If AppSignal is configured to start when the Rails application loads, rather than after it has initialised, only add Rails' instrumentation middlewares if AppSignal was actually started.

View the Ruby gem v4.7.0 changelog for more information.

MCP improvments, and Logs Updates

Added

• Add 3rd party integration links (e.g. GitHub, Jira) to MCP get_incident tool. The AI Agent can use this to tie incidents to external issues.
• Add uptime monitoring metrics to the MCP metric discovery tool
• Support ANSI colors in Logs page
• Persist the option to hide the logs chart between queries.

Logging updates

Added

• Add examples and “click to apply” functionality for IN and NOT IN in the logs search dropdown.
• Log sources now support automatic format detection. We will attempt to parse the line as JSON, Logfmt, and plain text, extracting attributes on every step.

Changed

• Hitting the escape key trying to close the log query help tool doesn't clear the query anymore.

Update OpenTelemetry dependencies

Changed

• Update the OpenTelemetry dependency to version 1.9 of the OpenTelemetry API. This improves compatibility with instrumentation libraries that depend on this version of the API.

View the Node.js package v3.7.1 changelog for more information.

Improved histograms and logs, report all child spans, and CORS support

Added

• Improve histogram support. When processing OpenTelemetry histograms, allow for histograms that do not have min, max or sum values. Use the exemplars provided alongside the histogram to improve the accuracy of the histogram.
• Return CORS headers from the export endpoints to make it possible to export OpenTelemetry data to these endpoints without the need of a proxy server. The Access-Control-Allow-Origin header value can be configured in the collector with the cors_origin config option or APPSIGNAL_CORS_ORIGIN environment variable.
• Support key-value list attributes in log lines. When an attribute sent in a log line is a key-value list, it will be converted into multiple attributes, one for each key-value pair in the list, prefixed with the key of the list attribute.

Changed

• Disable the default AppSignal request and response headers filter. By default, the OpenTelemetry SDK in the applications do not send any request and response headers and require configuration of to record these headers. This required double configuration in applications for AppSignal and OpenTelemetry. The appsignal.config.request_headers and appsignal.config.response_headers configuration options can still be used for additional header filtering if needed. Consult your application's language OpenTelemetry SDK for the configuration option to record request and response headers.
• Log a warning when making a GET request to a POST endpoint. The error page when making a GET request will also explain to make a POST request instead.
• Read the app language from the OpenTelemetry telemetry.sdk.language resource attribute if set. This attribute is set by the OpenTelemetry SDK for some languages, like Go, Java and PHP. If your application's OpenTelemetry SDK sets the telemetry.sdk.language resource attribute, you will not need to set the appsignal.config.language_integration attribute. If the appsignal.config.language_integration attribute is set, it will take precedence over the telemetry.sdk.language resource attribute.
• Report log lines with unspecified severity. When a log line does not have a severity, report it as a log line with info level severity.

Fixed

• Fix child spans not always being reported. This was due to the collector not properly traversing all the levels the trace span hierarchy.

This release can be installed through our collector packages and Docker image.

Requests to other instrumented apps will be reported as separate traces

Changed

• Disable the OpenTelemetry trace context propagation across applications. If multiple applications are instrumented with AppSignal for Node.js using OpenTelemetry, the propagation across applications could lead to application data being reported the wrong application in AppSignal.

View the Node.js package v3.7.0 changelog for more information.

MCP Updates

Added

• Add the create_incident_note tool for the MCP server. This allows AI Agents to comment on incidents.

  

Changed

• The update_incident MCP Tool now accepts severity as an argument (and sets the severity accordingly).

Add 'via' label for logbook items and fix terminated color for k8s overview hover

Added

• Add "via" label to logbook items if they were created through a 3rd-party source (e.g. an MCP server)

  

Fixed

• Fix terminated color when exit code of the container is 0. It should be green, as it has done it's job successfully.

  

Error causes for OpenTelemetry errors

Added

• Add error causes for OpenTelemetry errors

  

Download samples as JSON

Added

• Allow downloading samples as JSON
• Include Unassigned option in the assignee filter
• Add a callout to the instrumentation docs when the trace timeline has a few items

Fixed

• Fix visual bugs in the integration button dropdown.

Assignment improvements and more information on K8s node pages

Added

• You can now select an assignee on Enter, when there's only one filtered entry, on the incident list and show pages.
• Add a table on the Cluster monitoring node inspect page, showing all the pods related to the selected nodes

kubernetes and Anomaly updates

Added

• Add the ability to treat missing data point as 0 for performance and error triggers. If you have traffic in bursts (e.g. a Cron job), there's not always enough data to trigger or close an alert. When checking this checkbox, we will simulate the metric, allowing alerts to close.

  

• Add container status, uptime, and restart count in the Kubernetes Pods table
• Add container status and owner reference grouping options in the Kubernetes overview page
• Add uptime and restart count in the Pod overview tooltip

Changed

• Update the order of steps in the OpenTelemetry Java installation flow to improve the installation flow experience.

MCP Discover metrics & Kubernetes state pop-over

Added

• Add ability for Ai Agents to discover our metrics on our MCP server. Before this discover_metrics endpoint, we had our get_metric_names endpoint, but it was difficult for agents to determine what metric should be used when.

  With this new discover_metrics endpoint, we provide the Agent with a global overview of metric categories, and when they request more information about a category, we describe the metric names/tags/fields, and when to use them.

  We also take custom metrics and dashboards into account. Providing descriptive names for dashboards/description and charts on your dashboards helps the agent use those metrics.

Changed

• Improve cluster state pop-over. Show state of containers, if there are any.

  

Application Updates

Added

• Add get_metrics_timeseries endpoint to the MCP server. This endpoint exposes timeseries metrics to an AI Agent, enabling queries such as

  Show me the load average for all front-end machines on MyApp/production

  Shell

  Here's a summary of the load averages for the two frontend machines:
  
  1. frontend1:
    - Load average range: 0.83 - 1.6
    - Current load: 0.94
  
  2. frontend2:
    - Load average range: 2.0 - 3.01
    - Current load: 2.18
  
  The frontend-ams1 machine appears to have a significantly higher load compared to frontend1, with load averages consistently above 2.0, while frontend2 stays mostly below 1.5.
  

  We are still working on metric discovery. Right now you have to be quite specific to get the desired metrics.

• Add kubernetes volume metrics to node page.

  

Type signatures and improved Capistrano & Hatchbox deploys

Added

• Add Sorbet and RBS type signatures for the gem's public APIs. If your editor supports showing type signatures, they will now show up in for the AppSignal Ruby gem.
• When a deployment tool (Capistrano, Hatchbox.io) provides the REVISION file in the deployed application, AppSignal will now use that to set the revision config.

Changed

• Improve the YARD documentation for public APIs. This will make the documentation clearer and more useful for developers using the Ruby gem. Private APIs have been hidden from the generated output.
• When an error occurs while initializing AppSignal or when running a probe, use the error log level to log the error's backtrace.

Fixed

• When a Hash-like value (such as ActiveSupport::HashWithIndifferentAccess or Sinatra::IndifferentHash) is passed to a transaction helper (such as add_params, add_session_data, ...) it is now converted to a Ruby Hash before setting it as the value or merging it with the existing value. This allows Hash-like objects to be merged, instead of logging a warning and only storing the new value.

  Ruby

  # Example scenario
  Appsignal.add_params(:key1 => { :abc => "value" })
  Appsignal.add_params(ActiveSupport::HashWithIndifferentAccess.new(:key2 => { :def => "value" }))
   
  # Params
  {
    :key1 => { :abc => "value" },
    # Keys from HashWithIndifferentAccess are stored as Strings
    "key2" => { "def" => "value" }
  }

View the Ruby gem v4.6.0 changelog for more information.

Application MCP endpoint

Added

• Add MCP endpoint to list all applications a user has access to.
• Add Kubernetes overview page. This page shows your entire Kubernetes cluster state in one glance.