# Why Did AWS Ship MCP Apps Before Datadog or Grafana?

According to the AWS Machine Learning Blog, OpenSearch Service's MCP Apps allow an AI agent to return “interactive visualizations alongside your AI agent's text responses.” That is a subtle but decisive shift. Observability tools have spent a decade optimizing for human querying—dashboards, alerts, and manual drill-downs. AWS just re-architected the same data plane to serve an autonomous agent that can navigate from an alert to a trace to logs and propose a root cause in one conversation, with every step verifiable inline.

Why AWS? Because OpenSearch Service is the only major observability backend that is both open-source-derived and cloud-native, giving AWS the latitude to embed MCP directly into the query engine. Datadog and Grafana, by contrast, have proprietary data pipelines that would require deeper architectural changes. AWS's move is not just a feature; it is a statement that the agent is the new UI.

# What Does MCP Apps Change for the Observability Workflow?

The AWS blog describes a single, locally run MCP server as the connective tissue. That server links an agent to OpenSearch Service, allowing it to issue queries, receive structured results, and render those results as charts, traces, and log excerpts—all inside the agent's response. For a developer, that means asking “why did checkout latency spike at 14:32?” and getting not just a text explanation but a time-series graph and a trace waterfall, with the ability to drill into a log line without switching tools.

That is a workflow change, not a feature addition. The agent becomes a junior engineer that can show its work. AWS's documentation for MCP Apps (linked from the blog) confirms that the server runs locally, so the agent's context window carries the full lineage of the investigation. This is the first time an observability vendor has made verification a first-class agent output, not an afterthought.

# How Does MCP Apps Stack Against Datadog and Grafana?

The comparison is stark. Datadog and Grafana have invested heavily in AI assistants, but those assistants still operate in a separate pane, generating text and links. AWS has made the agent's output itself the visualization layer. That is not a minor UI difference; it is a philosophical one about who drives the investigation.

# Who Benefits Most from Agentic Observability—and Who Loses?

According to the AWS Machine Learning Blog, the entire loop—alert, trace, logs, root cause—happens “in one conversation, and you can verify every step inline without leaving your IDE.” That phrasing targets two audiences: developers who hate context-switching, and platform teams that want to reduce mean time to resolution (MTTR). The biggest winners are engineering teams already using OpenSearch Service or Amazon Managed Grafana, because they get agentic observability without a new vendor.

The losers are commercial observability vendors that have not yet embraced MCP. Datadog, in particular, has a moat built on integrations and dashboards, but if agents can query any data source via MCP, that moat erodes. Grafana Labs has an open-source advantage, but its agent story remains vague. AWS's move also pressures Splunk, which has been slower to adopt MCP in its observability portfolio.

# What Are the Risks and Limitations of This Approach?

No vendor ships a silver bullet. AWS's MCP Apps require that an agent can query OpenSearch Service effectively, which means the underlying data must be structured and indexed. The AWS blog does not discuss security controls for MCP Apps, but the local server model implies that access is governed by the user's existing IAM permissions. That is a reasonable start, but it also means that a compromised agent could exfiltrate sensitive traces or logs.

Another limitation is that interactive visualizations are only as good as the agent's ability to choose the right chart. AWS has not disclosed how the agent decides between a line chart and a histogram, and that could lead to misleading interpretations. The blog's example is a single spike; real incidents are messier. Still, the ability to verify inline mitigates the worst failure modes.

# What Should Teams Do Next?

For teams already on OpenSearch Service, the path is clear: spin up the MCP server, connect it to your IDE, and run a pilot where an agent investigates a known incident. The AWS blog provides the setup steps, and the documentation is live. For teams on Datadog or Grafana, the message is different: start asking your vendor about their MCP roadmap, because the agent era will not wait.

The broader lesson is that observability is becoming a conversation, not a dashboard. AWS has made the first credible move, and the incumbents' response will define the next phase of the market.