Agentic AI
Grafana Releases Live Telemetry Tools for AI Coding Agents to Verify Behavior
Grafana Labs has launched general availability for gcx and its MCP server, enabling AI agents to pull real observability data during development to ground code decisions in actual system behavior.
InfoQ
Grafana Labs has released two new tools to general availability that help AI coding agents access live observability signals while writing code. The gcx CLI and Grafana MCP server let agents query metrics, logs, traces, SLOs, and synthetic monitoring results from Grafana Cloud or self-hosted observability stacks. By connecting agent-generated code to real system behavior, these tools aim to close a confidence gap in automated development workflows.
The Problem with Unobserved Code Changes#
In traditional development, engineers build mental models of their changes as they write code. But when AI agents author pull requests, that gradual understanding evaporates—yet code review agents and automated approvals create a false sense of confidence anyway. Developers can merge PRs without truly grasping what the agent wrote or how it might affect production. Grafana's solution treats observability data as an additional, evidence-based verification layer rather than relying solely on code inspection.
Two Tools for Different Use Cases#
The Grafana MCP server provides opinionated, pre-configured tools for common tasks. It works either self-hosted or through a managed endpoint for Grafana Cloud users. The gcx CLI takes the opposite approach—it's lightweight and flexible, letting agents build custom workflows that work with any Grafana deployment, including open-source and enterprise versions. Grafana Labs ships installable skill bundles and Claude Code plugins for both, making integration straightforward.
Grounding Decisions in Real Data#
Instead of guessing how a system behaves, agents now check actual performance metrics before implementation. In Grafana's payment-provider example, an agent observes that the existing provider has a p95 latency of two seconds, then uses that figure to configure realistic mocks for unit and integration tests. This informs request-rate estimates for a new handler. Agents can also read and modify existing dashboards, trace queries back to their source code, and push updated definitions into Grafana or version control.
