Changelog

Follow us on LinkedIn
v0.121.1May 01, 2026

Introducing the SigNoz MCP Server

SigNoz now has an MCP server that lets AI assistants like Cursor, VS Code / GitHub Copilot, Claude, Codex, Gemini, and others query your observability data through natural language. You can search logs, query metrics, inspect traces, list alerts, and work with dashboards directly from your AI coding tool.

SigNoz Cloud users can connect to a hosted MCP endpoint directly from Settings, with client-specific setup steps and copyable configuration snippets. Self-hosted users can run the MCP server locally via binary, Docker, or by building from source. You can also explore ready-to-use workflows in the MCP use cases guide.

Better Graph Tooltips

Dashboard chart tooltips can now be pinned using a keyboard shortcut. While hovering over Time Series, Bar, or Histogram panels, users can press the P key to lock the tooltip in place and inspect values without keeping the cursor fixed on the chart.

The tooltip UI now shows clearer Live/Pinned states, contextual hints for available actions, and an Unpin action. The footer has also been polished to better distinguish drilldown and pinning interactions.

Deploy SigNoz Under a Custom URL Path

Self-hosted SigNoz deployments can now be served from a custom URL path, such as https://example.com/signoz/, instead of requiring a root domain. This gives teams more flexibility when deploying SigNoz behind reverse proxies, ingress controllers, or shared internal domains.

The external URL can be configured through the SIGNOZ_GLOBAL_EXTERNAL__URL environment variable, while the default remains unchanged for existing deployments. Once configured, SigNoz uses the custom path consistently across the application without requiring an image rebuild.

Please see the corresponding documentation for more details.

Bug Fixes

  • Dashboard Bar and Time Series panels now render correctly when groupBy data is missing, fixing crashes seen in panels created through LLM/MCP workflows.
  • Unsupported ClickHouse and PromQL query types have been disabled for Anomaly Detection alerts.
  • Several edge cases in the Alerts and Planned Downtime frontend flows, such as stale loading states, have been resolved.
  • Operations on deleted service accounts now return a proper 404 Not Found HTTP response instead of 501 Not Implemented.

Maintenance

  • Frontend internals were streamlined with remaining xstate usage replaced with plain React state and local dispatch helpers, light-mode specific styles cleaned up, and unused code deleted.
  • The Trace Waterfall API was moved into the Trace Detail module, with improvements to handling of large traces, caching, and error resilience.
  • Alerts and Planned Downtime frontend flows were migrated to generated clients, reducing custom API adapter code and improving consistency in error handling.
  • Integration and E2E test infrastructure was consolidated under a shared pytest project with common fixtures, local backend-based E2E runs, and updated CI workflows.
  • oxlint and oxfmt were added as frontend linter and formatter respectively, speeding up the developer feedback loop.
  • The OpenTelemetry Collector was bumped to stable v0.144.3.