Netdata AI can now connect outward to the tools your team already runs, like GitHub, PagerDuty, Atlassian, or any custom MCP server, and read from them during an investigation. We call this MCP Connections. It’s the missing piece in the middle of every root-cause investigation: the alert tells you what changed, but the why is usually somewhere else entirely.
The missing context
A latency spike fires an alert. The metrics show exactly when it started and which service it hit. They don’t show the deploy that went out four minutes earlier, the PagerDuty incident already open for the same symptom, or the runbook someone wrote for this exact failure mode six months ago. That context exists. It’s just scattered across a pull request, a ticketing system, and a wiki page, in tools your monitoring platform has never been able to see.
Up to now, that correlation was a manual job: alert in one tab, GitHub in another, PagerDuty in a third, Confluence in a fourth. Netdata AI could tell you what changed on the infrastructure, but it couldn’t see the pull request or the incident.
Netdata AI as an MCP client
This is the inverse of what we shipped earlier this year with the Netdata Cloud MCP Server, which let external AI assistants like Claude and Cursor query into Netdata. MCP Connections runs the other direction: Netdata Cloud itself is the MCP client, reaching out to the Model Context Protocol servers your team already operates and reading from them while it investigates.
That means Netdata AI can now pull external context, like a merged pull request or a tracked incident, alongside the metrics, logs, and anomalies it already collects from every node, and reason across both in a single pass. In practice:
- Tying a latency spike to the deploy that caused it
- Linking an anomaly to the incident that’s already tracking it in PagerDuty
- Surfacing the relevant runbook without anyone going to look for it
The investigation comes back with that context already included, so nobody has to correlate across five browser tabs.
What’s supported today
MCP Connections includes built-in providers for:
- GitHub (OAuth or personal access token)
- PagerDuty (incidents, on-call schedules, change events)
- Atlassian Cloud (Jira, Confluence, Bitbucket)
- Any custom MCP server reachable over HTTPS
Connecting a custom server means pointing Netdata at an HTTPS endpoint with a publicly trusted TLS certificate. Private and loopback addresses aren’t accepted. Requests time out at 45 seconds and tool results are truncated at 64 KiB, with authentication handled via a bearer token in the Authorization header.
Read-only, on purpose
Netdata AI can read from every connected tool but can’t act on any of them. Only read-only tools are ever enabled. A connected server may expose mutating actions, and Netdata will show them, but they can’t be turned on. Mutating actions are never available through Netdata.
Using connections
Once a connection is configured, you decide when Netdata AI uses it. In live conversations, you toggle connected servers on or off per conversation. For scheduled reports and automated investigations, you select which servers are available before the run happens. Either way, tool calls against connected servers don’t cost extra: an investigation is one AI credit on completion, regardless of how many external tools it queried along the way.
Getting started
MCP Connections requires a paid Netdata Cloud plan and Space admin access to configure. Once a connection is set up (bearer token or OAuth, depending on the provider), it’s available to every investigation and conversation in the Space, subject to the read-only constraints above.
Full setup details, including the exact requirements for pointing Netdata at a custom MCP server, are in the MCP Connections documentation.
We walked through this live in a recent webinar, Netdata AI Now Talks to Your Other Tools to Find Root Cause Faster, including a real investigation correlating a latency spike with the deploy that caused it. If you want to see the difference between an agent that reports signals and one that can explain them, it’s worth watching.