We’ve reviewed 10 categories of Observability & Monitoring MCP servers, evaluating over 75 individual servers across application performance monitoring, error tracking, incident management, log aggregation, uptime monitoring, and LLM observability. Each review covers architecture patterns, star counts, tool inventories, known issues, and honest ratings.
Observability is where MCP is seeing the strongest vendor adoption outside of cloud platforms. Grafana (2,600+ stars, 252K+ Docker Hub pulls) leads the ecosystem by a wide margin, while Datadog, Sentry, PagerDuty, Honeycomb, and New Relic have all shipped official first-party servers. The pattern is clear — observability vendors understand that AI agents need production context, and MCP is the interface they’re betting on.
Full-Stack Observability Platforms
The major observability platforms that provide unified dashboards, metrics, traces, and alerting. All five vendors in this group ship official first-party MCP servers — the highest vendor adoption rate of any subcategory.
| Review | Rating | Key Servers |
|---|---|---|
| Grafana | 4/5 | mcp-grafana (2,600+ stars, 303 forks, v0.11.3), 40+ tools across 15 categories |
| Datadog | 4/5 | Official Datadog MCP (22 stars, GA since March 9 2026), 50+ tools across 14 modular toolsets |
| New Relic | 4/5 | Remote-hosted at mcp.newrelic.com (Streamable HTTP), 35+ tools across 6 categories, Public Preview |
| New Relic NRQL & Alerts | 3.5/5 | Official first-party (4 stars, Public Preview), 15+ observability tools across 6 categories — NRQL queries, alerts, entity discovery, log analysis |
| Honeycomb | 4/5 | Official hosted MCP (replaced deprecated self-hosted, 41 stars, 12 forks), 14 tools, GA since Sep 2025, Agent Skills + Automated Investigations launched Mar 2026 |
Error Tracking & Incident Management
The tools teams reach for when things break — from error monitoring to on-call coordination. Both Sentry and PagerDuty ship official servers with deep write capabilities.
| Review | Rating | Key Servers |
|---|---|---|
| Sentry | 4/5 | sentry-mcp (603 stars, 93 forks, v0.30.0), ~20 tools |
| PagerDuty | 4/5 | pagerduty-mcp (57 stars, 30 forks, v0.15.1, 270 commits), 60+ tools |
Log Management & Uptime Monitoring
Log aggregation across Splunk, Elasticsearch, Loki, and CloudWatch, plus uptime and status monitoring with UptimeRobot, Uptime Kuma, and Better Stack.
| Review | Rating | Key Servers |
|---|---|---|
| Log Management | 4/5 | 25+ servers across 12 platforms — mcp-grafana leads (2,500 stars, 6 Loki tools), Elasticsearch official (626 stars, deprecated), Splunk official + community (94 stars, 13 tools), CloudWatch, Graylog |
| Monitoring & Uptime | 3.5/5 | 10+ servers across 5 platforms — UptimeRobot (official hosted), Uptime Kuma (4+ community servers), OneUptime (711 endpoints across 126 resource types), Better Stack, Checkly |
LLM Observability & MLOps
The emerging subcategory where observability meets AI — tracking LLM traces, prompt performance, model evaluation, and ML pipeline orchestration.
| Review | Rating | Key Servers |
|---|---|---|
| LLM Observability & MLOps Pipeline | 3.5/5 | 10+ servers across 5 subcategories — Opik (200 stars, LLM traces/evals), OpenTelemetry MCP (175 stars, Jaeger/Tempo/Traceloop), Langfuse (158 stars, prompt management), LangSmith (89 stars), ZenML (43 stars, pipeline orchestration) |
Category Overview
10 reviews. 75+ servers. Average rating: 3.9/5.
What stands out
Vendor adoption is extraordinary. Grafana, Datadog, Sentry, PagerDuty, Honeycomb, and New Relic all ship official first-party MCP servers. That’s six major observability vendors — a higher official adoption rate than almost any other MCP category. The message is clear: observability companies see AI agents as first-class consumers of production telemetry.
Grafana is the runaway star count leader. At 2,600+ stars and 252K+ Docker Hub pulls, mcp-grafana dwarfs every other observability MCP server. Its 40+ tools across 15 categories cover dashboards, alerting, Loki logs, Prometheus queries, and more. It’s also the backbone of the log management ecosystem, providing the most mature Loki integration.
Sentry and PagerDuty prove write-capable MCP is production-ready. Sentry’s 603-star server lets agents debug production errors without leaving the editor. PagerDuty’s 60+ tools make it the most comprehensive write API in the incident management space — agents can create, acknowledge, and resolve incidents, manage on-call schedules, and coordinate response teams.
Honeycomb pioneered the hosted-MCP model. When Honeycomb deprecated its self-hosted server in favor of a hosted MCP endpoint, it became one of the first observability vendors to treat MCP as a managed service. The March 2026 launch of Agent Skills and Automated Investigations suggests the future: vendors won’t just expose data through MCP — they’ll ship autonomous investigation capabilities on top of it.
LLM observability is the newest frontier. With Opik, Langfuse, LangSmith, and OpenTelemetry MCP all shipping MCP servers, a dedicated ecosystem is forming around tracking LLM traces, evaluating prompt performance, and orchestrating ML pipelines. This subcategory barely existed six months ago — it’s now one of the fastest-growing areas in MCP.
Sources
- grafana/mcp-grafana — Official Grafana MCP server (2,600+ stars, 40+ tools)
- datadog-labs/mcp-server — Official Datadog MCP server (GA March 2026)
- newrelic/mcp-server — New Relic MCP server (remote-hosted, Public Preview)
- honeycombio/honeycomb-mcp — Honeycomb MCP server (deprecated self-hosted, succeeded by hosted endpoint)
- honeycombio/agent-skill — Honeycomb Agent Skills (launched March 2026)
- getsentry/sentry-mcp — Sentry MCP server (603 stars, 20 tools)
- PagerDuty/pagerduty-mcp-server — PagerDuty MCP server (60+ tools)
- comet-ml/opik-mcp — Opik LLM observability MCP server
- traceloop/opentelemetry-mcp-server — OpenTelemetry MCP server
- langfuse/mcp-server-langfuse — Langfuse prompt management MCP server
- langchain-ai/langsmith-mcp-server — LangSmith MCP server
- zenml-io/mcp-zenml — ZenML pipeline orchestration MCP server