Ask Questions in Your Language, Get Answers From Your Infrastructure

MCP is an open protocol enabling AI assistants to access external data sources through standardized interfaces. For infrastructure monitoring, MCP transforms AI assistants from generic chatbots into infrastructure experts - they can query real-time metrics, analyze ML-detected anomalies, search logs, and access live system state. Netdata’s built-in MCP server eliminates the gap between asking a question and getting an answer from your infrastructure.

Netdata ships MCP as a built-in capability in every Agent and Parent (v2.6.0+) - no external server installation, no configuration required. Competitors require separate MCP server deployment, extensive configuration, or are in preview/beta status. Netdata’s MCP includes unique features competitors don’t offer: real-time ML anomaly detection embedded in every metric sample, direct logs access without pipelines, and console replacement capabilities (live processes, network connections, services).

All MCP-compatible clients work with Netdata: Claude Desktop, Claude Code, Cursor IDE, ChatGPT (via Cursor), Gemini CLI, VS Code (with MCP extensions), JetBrains IDEs, and any custom client using MCP protocol. Netdata supports multiple transports (streamable HTTP, SSE, WebSocket, stdio with a bridge) for maximum compatibility. Setup takes under 5 minutes - install Netdata, grab the auto-generated API key, add to AI assistant config.

No for basic metrics queries - MCP works on standalone Agents and Parents. Yes for sensitive functions (logs, processes, network connections) - production use requires claimed agent to Netdata Cloud for authorization. This ensures secure, auditable access while maintaining data sovereignty. All observability data stays on-premises; only metadata synchronizes to Cloud for authentication.

MCP tools provide comprehensive access: discover nodes/metrics/functions, query time-series with ML anomaly rates, find correlated/anomalous/unstable metrics for root cause analysis, list/query alerts with complete transition history, execute live functions (processes, network connections, systemd-journal, Windows events), and access complete logs without pipelines. AI assistants get same capabilities human users have, plus natural language interface.

MCP is included in all Netdata pricing tiers at no additional charge - even the free Community plan. Business plan includes unlimited MCP queries, unlimited AI assistant access, and unlimited users with predictable per-node pricing. No per-query charges, no data volume fees, no feature gating. This is fundamentally different from competitors who charge based on data volume or queries - Netdata’s flat per-node pricing encourages unlimited AI usage.

Yes - designed with security-first architecture: read-only access model (no configuration changes possible), Bearer token authentication (API key required), claimed-agent requirement (production use requires Netdata Cloud authorization), no sensitive data exposure (passwords, secrets, API keys protected), complete audit trail (all queries logged), and data sovereignty (metrics/logs stay on-premises). SOC 2 Type 2 certified with GDPR/HIPAA alignment by design.

Same way Netdata scales - through distributed architecture. Query single Agent for single-node data, query Parent for aggregated multi-node data, or use Netdata Cloud to distribute queries across all Agents/Parents. AI assistants see unified namespace regardless of infrastructure size (1 to 100,000+ nodes). No architectural changes required as you scale - same MCP interface works at any size.

Two distinct but complementary capabilities: MCP Server is a built-in protocol adapter enabling ANY MCP-compatible AI assistant to query Netdata data - bring your own LLM (Claude, ChatGPT, Gemini, etc.), free with open-source Netdata, included in all Cloud plans. Netdata AI Features are managed AI services (AI Insights, AI Chat, AI Troubleshoot) using Netdata-optimized playbooks - requires Netdata Cloud plus AI credits (10 free/month in Business plan). Many customers use both: MCP for ad-hoc developer troubleshooting, AI features for automated reports and management visibility.

Yes - Netdata Agents/Parents with built-in MCP servers work completely offline. AI assistants connect directly to local Netdata instances via streamable HTTP, SSE, or WebSocket. For multi-node visibility in air-gapped environments, deploy on-premises Netdata Cloud or use Netdata Parents. No internet connectivity required for MCP functionality - all queries execute locally against edge-native data.

Three steps taking under 5 minutes: (1) Install Netdata Agent/Parent (already includes MCP server, no separate installation), (2) Get API key from the Netdata configuration directory, (3) Add to AI assistant config with endpoint http://your-netdata:19999/mcp. Most teams are asking infrastructure questions within 5 minutes. See detailed setup guide for specific AI assistant configurations.

APIs are designed for machines, not AI reasoning. MCP provides semantic context - AI understands what metrics mean, not just their values. Netdata’s MCP includes ML anomaly detection in every sample, so AI distinguishes normal from abnormal behavior automatically. Traditional APIs require AI to interpret raw values without context, leading to incorrect conclusions. MCP also provides standardized tool discovery - AI assistants automatically understand available capabilities without custom integration code.

Anomaly detection embedded in every metric sample transforms AI understanding. Instead of seeing just values, AI sees [value, anomaly_rate_%, annotations] - understanding which metrics are behaving abnormally without manual threshold definitions. 18 models per metric train continuously with 99% false positive reduction through consensus. This enables AI to provide accurate root cause analysis in the top 30-50 correlated metrics, identify cascading failures, and suggest remediation based on actual abnormal behavior - not just threshold violations.

Yes - Netdata MCP server handles concurrent connections from multiple AI assistants without conflicts. Each connection is independent with separate authentication. Multiple team members can use different AI assistants (Claude, Cursor, ChatGPT, etc.) simultaneously without performance degradation. This enables team collaboration where different engineers use their preferred AI tools while accessing the same infrastructure data.

Valid option for metrics-only access, but Netdata provides significantly more: ML anomaly detection (real-time, 18 models/metric, embedded in samples - Prometheus doesn’t have), logs access (direct systemd-journal/Windows Event Log querying - Prometheus doesn’t do logs), console functions (live processes, network connections, services - Prometheus doesn’t provide), zero configuration (built-in MCP server - Prometheus requires external MCP server setup), automated dashboards (algorithmic organization - Prometheus requires Grafana setup), and enterprise support (official product with SOC 2 Type 2 certification - community MCP servers unsupported).

Five common mistakes: (1) Not claiming agents - production use requires claiming for authorization, (2) Forgetting API key security - treat API keys as credentials, don’t commit to git, (3) Overwhelming AI with broad queries - start specific (‘show CPU on node-x’) before broad (‘analyze entire infrastructure’), (4) Ignoring anomaly context - train team to ask AI about anomaly rates, not just values, (5) Not using specialized tools - use find_correlated_metrics for root cause analysis instead of manual correlation. Following best practices accelerates adoption and maximizes value.

Effective query patterns: Discovery first (‘List all nodes’ → ‘Show metrics for node-x’ → ‘Query CPU metrics’), Use context (‘Show anomalous metrics in last hour on production nodes’), Use ML (‘Find metrics with highest anomaly rates during incident at 2pm’), Combine data types (‘Show logs for services with raised alerts’), Time-bound queries (always specify time windows for metric queries), Filter appropriately (use labels, patterns to narrow large infrastructures). These patterns help AI assistants provide accurate, actionable insights.

No - Netdata’s MCP server provides data access only. YOUR chosen AI assistant (Claude, ChatGPT, Gemini, etc.) sends queries to Netdata and receives responses. What the AI provider does with that data depends on your agreement with them (many offer enterprise plans with no data retention). Netdata itself never sends your observability data to AI providers. This architecture maintains complete control over data sharing - you choose which AI provider to trust.

Yes - all MCP queries are logged with timestamps, source IPs, authenticated users, and queries executed. Netdata Cloud provides centralized audit logs across all Agents/Parents. For compliance, export audit logs to SIEM systems. API key-based authentication enables tracking which AI assistants/users accessed what data when. This complete audit trail satisfies regulatory requirements and enables security team oversight of AI assistant usage.

Negligible - MCP server adds approximately 5KB RAM per active session and less than 0.1% CPU. Netdata’s architecture processes queries from existing in-memory data structures without additional database load. Even with multiple AI assistants querying simultaneously, impact remains minimal due to efficient C implementation and read-only access model. The University of Amsterdam study validated Netdata as the most energy-efficient monitoring solution - MCP capability maintains this efficiency.