SigNoz MCP Server

SigNoz is the open-source, OpenTelemetry-native observability stack — metrics, traces and logs in one ClickHouse-backed product — and this is its official MCP server, written in Go and shipped by the SigNoz team under Apache-2.0.

It is a read-and-write server, not a read-only bridge. On the read side there are tools for discovering active metrics and their cardinality, running Query Builder v5 queries, listing APM services and their top operations, aggregating or searching logs and traces, pulling a full trace with its span hierarchy, and reading alert instances, alert rules, dashboards and saved Explorer views. On the write side it can create, update, delete and JSON-Patch dashboards, create and replace alert rules (v2 direct and policy-routed, plus v1 anomaly rules), manage notification channels and save Explorer views. Two extra tools search and fetch SigNoz's own documentation, so an agent can answer "how do I ship Docker logs here?" without leaving the session.

Some details make it noticeably better behaved than a generic API wrapper. Read tools return a webUrl deep link to the matching page in the SigNoz UI when an instance URL is available, so an answer comes with somewhere to click. Argument validation never hard-fails a call: a schema mismatch still runs best-effort and appends a notice naming the offending parameter. Upstream 401 and 403 errors carry a structured UNAUTHORIZED or PERMISSION_DENIED code with recovery guidance rather than a raw passthrough. Bundled MCP resources document alert and dashboard schemas, widget choices and JSON-Patch recipes, which is what keeps agent-authored dashboards valid.

SigNoz Cloud hosts it at https://mcp.<region>.signoz.cloud/mcp with no install — pick the region matching your account. Self-hosted deployments run the published Docker image against their own SigNoz instance. There is also a packaged Claude Desktop extension. Protocol support covers the 2025-11-25 and 2026-07-28 MCP revisions.