Bringing Agent-Native Observability to SigNoz

Most observability workflows still assume one user: an engineer in a dashboard, scanning panels, scrolling logs, writing queries, and creating alerts.

But now, there is a second user - AI agents.

Agents do not use observability tools like humans do. They work through APIs, schemas, natural-language descriptions, and raw telemetry, which makes them good at correlating signals and reporting back quickly. They still need humans to decide which findings matter.

What happens when you give them access to your observability data? How do humans work with agents to build more resilient applications?

At SigNoz, we believe agent-native observability marks an important shift in how teams work with production data. We are investing in that direction, starting where software gets built today: inside your coding agent.

Enabling agent-native observability

The first step for agents is to interact with observability data. And, we're excited to announce that the hosted SigNoz MCP server is now live for all SigNoz Cloud users. The hosted Model Context Protocol server connects SigNoz to MCP-compatible agents, including Claude Code, Cursor, Codex, Gemini CLI, and custom agents. Cloud users get it without installing anything; self-hosted teams can run their own open-source MCP server themselves. The agent can ask, in natural language, what alerts fired in the last hour, which services degraded after the 14:03 deploy, what this customer's last session actually did, and get answers grounded in your real telemetry, not a generic LLM hunch.

And it doesn’t stop there.

Bringing observability into your coding agents through MCP enables interfaces that were earlier not possible. You can use AI agents to connect dots using data from different aspects of your business and get intelligent insights. One customer rightly put it, “This goes well beyond SigNoz.”

A surprise use case we saw with one of our customers is that our MCP server made observability data easier for non-engineering teams to use. It is now being used by teams like customer support to gather context that helps them serve their users better.

Apart from our MCP server, we’re also working on a few other things that will help you use AI across observability workflows. Our AI Assistant, currently in beta, will introduce a conversational layer inside the SigNoz UI, letting you explore data quickly and build dashboards and alerts from the chat interface itself.

Foundations of good observability start with what data you collect. We’re also working actively on shipping skills that will help you instrument your applications well. AI agents can learn your team's conventions, rules, and patterns that will then enable coding agents to instrument new code the way your team would. Which metrics matter, which fields explode the bill, which tracing patterns hold up at scale.

With SigNoz bringing AI into observability workflows, you can focus more on things that matter for your users. But this brings up a question we keep getting asked.

So, are humans still needed for observability in the agentic era?

Yes. And here's why.

An agent with access to SigNoz has more context than any human ever has. It can read every trace. It can scan every log. It can correlate ten metrics across five services in a hundred milliseconds. The context problem in observability — "I don't have time to look at all of this" — is, for the first time, getting solved.

But context isn't the same as care. Agents can find patterns the moment they happen, but they don't know which patterns matter to your business — and what matters keeps evolving.

Because humans know what to care about, critical workflows need their input and stewardship.

We've watched teams try to give that judgment to an agent. The result is the same every time: the agent generates noisy alerts, the team learns to ignore them, and within a couple of weeks, the experiment is uninstalled. As one engineer told us: "If it once gives you a false alert, you just start ignoring it." Without a human who cares about the feature, an alert isn't a signal — it's noise that happened to clear a threshold and might not be a cause of concern.

So our view of what observability looks like in an agent-native world has two halves. Agents should gather evidence, correlate telemetry, and propose causes. Humans should set alert policy, choose what deserves a page, and decide when a pattern matters. The roles haven't disappeared. They've moved up the stack.

Why SigNoz fits the agentic era

SigNoz at its core is open.

Open-source, based on open standards like OpenTelemetry, and built to serve as a one-stop observability platform. We wanted to give all engineering teams access to a powerful observability stack without getting vendor-locked, or hassles of dealing with fragmented tooling. With AI agents now part of how engineering teams work, we are extending that mission to agents too.

Those choices now make SigNoz a strong foundation for agent workflows: OpenTelemetry, open source code, and one schema across logs, metrics, and traces. AI agents have good context of how OpenTelemetry works from thousands of well-documented community–led tutorials across the internet. At the same time, agents have access to our entire codebase to learn how this data is processed, and saved from different observability use-cases.

SigNoz also brings logs, metrics, and traces into a unified data store with a consistent schema. For agents, this matters more than for humans. An agent investigating a slow checkout can pull the trace, the matching logs from that span, and the metric on the dependency without stitching across three tools or normalizing three different data models. Schema consistency, in an agent-native world, is product quality.

We believed in reducing the context required for troubleshooting performance issues from day one. With AI agents, that mission now extends to them too. Agents can use the same unified telemetry to surface insights faster for the end user.

What's next

Over the next few months, we’re building deeper integrations with coding agents, adding more tools to the SigNoz MCP server, and extending agent workflows to more parts of observability.

We’re focused on three areas:

• Skills for instrumentation, dashboards, and team-specific debugging conventions, so your debugging knowledge can be reused across the team.
• Schema-aware suggestions inside SigNoz that flag drift between traces, metrics, and logs, so your agent does not have to infer the schema from scratch every time.
• Investigation-to-policy workflows that turn a debugging session into the next operational step. For example: “We just debugged this. Turn it into an alert, lower the threshold so I can verify it, and ship it.”

Our larger goal has not changed: make open observability accessible and useful for teams building resilient applications. What has changed is the shape of the team. It now includes humans and agents. SigNoz is being built for both: agents that gather context from telemetry, and humans who decide what matters.

To explore the full picture, head to Agent Native Observability in SigNoz. To connect your coding agent today, start with the SigNoz MCP server docs.