Which MCP Servers Connect to Monitoring and Observability Platforms?

Monitoring servers connect AI agents to the platforms where teams track application health, investigate incidents, and analyze system performance. Independent evaluations of every significant server in this category, covering Datadog, Sentry, Grafana, Prometheus, PagerDuty, and Splunk, not directory listings or vendor documentation.

Model Context Protocol (MCP) lets AI clients query metrics, search logs, triage errors, manage incidents, and render dashboards through a standardized JSON-RPC 2.0 interface. Anthropic introduced MCP as an open standard in November 2024. The Linux Foundation’s Agentic AI Foundation (AAIF) now governs the specification (Linux Foundation, “Agentic AI Foundation Launch,” December 9, 2025). 6 servers compete in this category as of mid-2026. Monitoring has the highest first-party vendor adoption rate of any MCP category: 5 of 6 servers are official vendor implementations.

The full MCP server directory covers all 14 server categories. This page focuses on monitoring and observability servers alone.

What Does a Monitoring MCP Server Do?

A monitoring MCP server exposes metrics querying, log searching, error triage, incident management, and dashboard rendering as callable tools. The AI client investigates alerts, correlates logs with traces, identifies root causes, and coordinates incident response through a standardized JSON-RPC interface.

Monitoring servers connect to the observability platforms teams already pay for. The practical advice for this category is simple: install the MCP server for the observability platform the team already uses. Datadog teams install Datadog MCP. Grafana teams install Grafana MCP. Sentry teams install Sentry MCP. Switching platforms is not required.

4 observability functions define this category:

• APM (Application Performance Monitoring) platforms connect to Datadog and similar services. Tools expose metrics, traces, logs, monitors, dashboards, and infrastructure data across the full application stack.
• Error tracking connects to Sentry. Tools expose issue search, release-aware stack traces, root cause analysis, and error grouping with deployment context.
• Open-source observability connects to Grafana and Prometheus. Tools expose PromQL (Prometheus Query Language) queries, Loki log searches, dashboard management, and multi-backend data access.
• Incident management connects to PagerDuty. Tools expose on-call schedules, incident creation, escalation policies, and response coordination.

Azure SRE Agent (GA April 2026) and AWS DevOps Agent (GA April 2026) both consume monitoring MCP servers as their primary investigation interface. The cloud providers validated this category by building production agents that depend on observability MCP connections. The evaluation criteria in the next section help teams select the right server from this validated ecosystem.

How Do You Evaluate Monitoring Servers?

Evaluate monitoring servers on 5 criteria: observability function, deployment model, auth security, tool surface breadth, and backend flexibility. These dimensions determine whether a server fits the team’s existing observability stack.

Observability Function

APM servers provide full-stack visibility: metrics, traces, logs, infrastructure. Error tracking servers specialize in application failures with release context. Open-source servers provide backend flexibility. Incident management servers coordinate response workflows. Match the function to the team’s operational needs.

Deployment Model

Sentry provides a zero-install remote server at mcp.sentry.dev via Streamable HTTP. Datadog provides an official remote MCP server. Grafana runs locally via uvx mcp-grafana. PagerDuty runs locally. Remote servers require no local installation. Local servers require runtime dependencies and configuration files.

Auth Security

Sentry uses OAuth 2.0 with no tokens stored on disk. Datadog uses API and application keys. Grafana uses service account tokens. PagerDuty uses API keys. OAuth-based servers (Sentry) provide per-user permission boundaries without credential files. Token-based servers require secure storage of credentials in configuration.

Tool Surface Breadth

Datadog exposes 50+ tools across 10+ modular toolsets. Grafana exposes 40+ tools across 15+ configurable categories. PagerDuty exposes 60+ tools. Sentry focuses on error tracking with fewer but deeper tools. Broader surfaces consume more context window tokens.

Backend Flexibility

Grafana connects to Prometheus, Loki, ClickHouse, CloudWatch, Elasticsearch, InfluxDB, and Graphite through one server. Datadog connects to Datadog’s proprietary backend only. Sentry connects to Sentry’s backend only. Backend flexibility determines whether one server covers the team’s full observability stack.

The rubric above applies to every server profiled in the next section.

Which Servers Connect to Monitoring Platforms Through MCP?

6 monitoring servers expose MCP-compatible tools as of mid-2026. Datadog provides the broadest toolset with 50+ tools. Sentry provides the cleanest auth model with zero-install OAuth. Grafana provides the most backend flexibility with support for 7+ data sources through one server.

Datadog MCP (Official)

Datadog maintains the official MCP server. Auth: API key + application key. Transport: remote (official) or local via npx @datadog/mcp-server. The server exposes 50+ tools across 10+ modular toolsets. Coverage spans APM (Application Performance Monitoring), infrastructure metrics, log management, error tracking, feature flags, LLM (Large Language Model) observability, database monitoring, and synthetic tests.

The most frequently called tool is get_monitors with a status filter for active alerts. Natural language queries translate to Datadog’s metrics DSL (Domain-Specific Language): “show me all 500 error rates for the API server over the last 10 minutes” executes the correct metric query. Token-budget pagination keeps responses within context window limits. GA status. HIPAA (Health Insurance Portability and Accountability Act) eligible.

Datadog MCP is the enterprise choice for teams already paying for Datadog. The server connects to Datadog’s proprietary backend only. Teams using self-hosted Prometheus and Grafana need the Grafana MCP server instead.

Sentry MCP (Official)

Sentry maintains the official MCP server at mcp.sentry.dev. Auth: OAuth 2.0 (no tokens on disk, no credential files). Transport: remote via Streamable HTTP. Zero local installation required. 671 GitHub stars and 981 commits.

Sentry MCP specializes in application error tracking. Seer AI (Artificial Intelligence) root cause analysis identifies why errors occur. Release-aware stack traces show which deployment introduced the failure. Issue grouping consolidates duplicate errors. The server answers “what broke and in which release” more precisely than any APM platform’s error view.

Sentry complements Datadog. Most teams run both: Datadog for infrastructure and APM observability, Sentry for application error monitoring. The MCP protocol allows both servers in the same client session. The agent queries Datadog for infrastructure metrics and Sentry for error details without switching tools.

Grafana MCP (Official)

Grafana Labs maintains the official MCP server with 2,900+ GitHub stars. Auth: service account token. Transport: local via uvx mcp-grafana. The server exposes 40+ tools across 15+ configurable categories: dashboard management, Prometheus queries, Loki log searches, ClickHouse queries, CloudWatch metrics, Elasticsearch searches, InfluxDB queries, and Graphite data.

Multi-backend support is the key differentiator. One Grafana MCP server connects to 7+ data sources. Teams running Prometheus for metrics, Loki for logs, and ClickHouse for analytics access all 3 through the same MCP connection. Dashboard PNG rendering enables visual output in agent conversations. Admin RBAC (Role-Based Access Control) tools manage user permissions.

Grafana MCP is the open-source alternative to Datadog. Teams self-hosting their observability stack with Prometheus and Loki choose Grafana MCP over Datadog MCP.

Prometheus MCP (Community)

Community-maintained Prometheus MCP servers translate natural language queries into PromQL and execute them against Prometheus instances. Auth: Prometheus endpoint URL (unauthenticated within private networks, token-based when exposed externally). Transport: local (stdio). Tools expose metric queries, alert rule inspection, target health checks, and label discovery.

PromQL has more training data coverage in LLMs than Datadog’s metrics DSL. Models generate PromQL queries more accurately than proprietary query languages. Teams running standalone Prometheus without Grafana install this server. Teams already using Grafana access Prometheus data through the Grafana MCP server instead, which provides PromQL access alongside Loki and other backends.

PagerDuty MCP (Official)

PagerDuty maintains the official MCP server with 60+ tools. Auth: API key. Transport: local. Tools expose on-call schedule queries, incident creation, status updates, escalation management, service listings, and response coordination. The server also exposes change events, maintenance windows, and team management operations.

PagerDuty MCP pairs with Datadog or Grafana MCP for end-to-end incident workflows. The agent detects an anomaly in Datadog metrics, correlates it with Sentry error data, identifies the root cause, and creates a PagerDuty incident with full investigation context attached. The incident routes to the correct on-call responder automatically.

Splunk MCP (Official, GA)

Splunk (Cisco) maintains the official MCP server at GA status. Auth: Splunk access token. Transport: remote. Tools expose log search, dashboard access, and security event analysis.

Splunk MCP targets enterprise teams using Splunk for log management and security operations. The server complements SIEM (Security Information and Event Management) workflows by enabling AI agents to search and correlate security events.

Server selection follows the existing observability stack. Datadog teams install Datadog MCP. Grafana teams install Grafana MCP. Sentry teams install Sentry MCP. PagerDuty adds incident management to any combination.

The sections above define monitoring MCP servers, provide an evaluation rubric, and profile every significant server in the category. The sections below cover multi-server observability stacks and answer the questions that surface during platform selection.

Which Servers Work Together for Full-Stack Observability?

Most teams run 2 monitoring servers: one for infrastructure/APM visibility and one for error tracking or incident management. Datadog + Sentry covers commercial stacks. Grafana + Prometheus covers open-source stacks. Add PagerDuty to either combination for incident response.

Commercial stack: Datadog MCP for metrics, logs, traces, and infrastructure. Sentry MCP for application errors with release context. PagerDuty MCP for incident escalation. The combined tool definitions consume approximately 4,000 to 7,000 tokens at session start.

Open-source stack: Grafana MCP for dashboards, Prometheus metrics, and Loki logs. Prometheus MCP adds standalone PromQL access for teams without Grafana. PagerDuty MCP coordinates incident response. The questions below address the selection decisions that follow stack configuration.

Frequently Asked Questions

Do I Need to Switch Observability Platforms to Use MCP?

No. Every major observability vendor now maintains an official MCP server. Install the server for the platform the team already uses. Datadog customers install Datadog MCP. Grafana users install Grafana MCP. Sentry users install Sentry MCP. No platform migration is required.

Why Does Sentry Use OAuth While Others Use API Keys?

Sentry’s zero-install remote server at mcp.sentry.dev authenticates through OAuth 2.0 in the browser. No API key touches the filesystem. Datadog and Grafana run locally and store credentials in configuration files. The auth model difference reflects deployment architecture: remote servers use OAuth because no local process stores credentials. Local servers use tokens because the process needs persistent access. 

Can Monitoring Servers Write Data or Only Read?

Most monitoring servers are read-heavy. Datadog MCP reads metrics and logs. Sentry MCP reads errors and traces. Grafana MCP reads dashboards and data sources. Write operations exist for incident management: PagerDuty MCP creates and updates incidents. Datadog MCP can mute monitors and manage dashboards. Read-only access is the default starting point for observability.

What Is the Minimum Observability Stack?

One server covers most workflows. Teams using Datadog install Datadog MCP. Teams using the open-source stack install Grafana MCP (which covers Prometheus and Loki through one server). Add Sentry MCP for application error tracking. Add PagerDuty MCP for incident response. The maximum practical stack is 3 servers: APM + errors + incidents.