{"slug": "grafana-s-gcx-and-mcp-server-reach-ga-for-telemetry-driven-agent-development", "title": "Grafana's gcx and MCP Server Reach GA for Telemetry-Driven Agent Development", "summary": "Grafana Labs announced general availability for the gcx CLI and Grafana MCP server, tools that let AI coding agents query live observability data from Grafana Cloud or self-hosted stacks during development. The tools, first announced at GrafanaCon 2026 in April, aim to give agents evidence-based checks on their output by pulling metrics, logs, traces, SLOs, and Synthetic Monitoring results, addressing the gap between agent-generated code and engineer understanding. Grafana Labs demonstrated workflows where agents use production telemetry to set realistic test parameters and generate load-test scripts, reducing manual effort from about a day.", "body_md": "Grafana Labs has [announced](https://grafana.com/blog/telemetry-driven-development-how-to-gain-confidence-in-your-coding-agents-behavior-with-gcx-and-grafana-mcp/) general availability for two tools that let AI coding agents query live observability data during development: the gcx CLI and the Grafana MCP server. Both allow agents to pull metrics, logs, traces, SLOs, and Synthetic Monitoring results from [Grafana Cloud](https://grafana.com/products/cloud/) or a self-hosted stack. This integration helps bridge the gap between the speed at which agents generate code and engineers' understanding of what that code does. These tools were announced at GrafanaCon 2026 in April, as [reported](https://www.infoq.com/news/2026/04/grafana-loki-ai-agents/) by InfoQ.\n\nThe tools address a common problem in agentic coding workflows. Developers can plan, implement, and review changes mostly using an LLM. However, they might still merge a PR without fully understanding it. Historically, engineers built a mental model of a change as they typed it. When an agent writes the diff instead, that model-building step disappears, even as review agents and \"LGTM\" approvals create a false sense of confidence. Grafana Labs frames its answer as pulling observability data into the loop as an additional, evidence-based check on agent output, rather than relying solely on code review.\n\nArchitecturally, the two tools serve different levels of opinionation. The Grafana MCP server exposes a fixed set of tools for common use cases and is available either self-hosted or through a hosted endpoint for Grafana Cloud stacks. gcx is a flexible choice. It’s a CLI that agents can use to create custom workflows. These workflows work with Grafana Cloud or self-hosted OSS and Enterprise instances. Grafana Labs also ships an installable skill bundle for gcx (gcx agent skills install) and separate Claude Code plugins for the MCP server and for a Grafana Assistant guidance layer.\n\nOn Reddit, [dafydd-t](https://www.reddit.com/user/dafydd-t/), who works at Grafana Lab, [said](https://www.reddit.com/r/grafana/comments/1v928us/gcx_and_grafana_mcp_server_are_now_generally/):\n\nThe Grafana MCP server provides more opinionated tools for common tasks when interacting with Grafana. gcx provides a broader, less opinionated set of tools that you can use to build your own workflows.\n\nIn practice, the workflow lets an agent ground implementation decisions in observed system behavior rather than assumptions from its training data. Grafana Labs shows how to add a new payment provider. Instead of guessing the load, the agent checks current RED metrics. It finds that the p95 latency for the existing provider is two seconds. Then, it uses this data to set a mocked latency for unit and integration tests. This helps estimate request rates for the new handler. Agents can read and change existing dashboards. They can trace dashboard queries to the code that generates the telemetry. Also, they can push updated dashboard definitions back into Grafana or source control.\n\nFor local iteration, Grafana Labs points to agents' existing familiarity with [OpenTelemetry](https://opentelemetry.io) tooling: agents can stand up an OpenTelemetry Collector and export local build telemetry to a Grafana stack, or use the [grafana/otel-lgtm Docker image](https://hub.docker.com/r/grafana/otel-lgtm) for a fully local LGTM stack. Gcx is configured to pull production dashboard definitions into that local instance via gcx. resources pull. Traffic generation works in a similar way. Agents can use production telemetry to create realistic k6 load-test scripts. They can also set up k6 in a container. Grafana Labs says this used to take about a day of manual work. k6 ships its own agent-skill bundle, installable via k6 x agent init, which can author tests from observed traffic and help distinguish a failing test from a failing system under test.\n\nGrafana Labs shares an internal example: the [Tempo](https://grafana.com/docs/tempo/latest/) squad uses a custom agentic harness. This harness profiles a development environment with several terabytes of data. It identifies candidates for optimization. After implementing changes, it runs the same baseline query to measure the difference. Then, a person decides whether to merge the changes. Grafana Labs has launched an experimental Agentic Testing feature. It uses natural-language instructions to check UI flows in a live web application. This addresses the issue that backend telemetry doesn't capture front-end regressions.\n\nGrafana Labs explains that a pull request (PR) now looks different. Instead of just an agent-generated diff and description, a PR can include links to dashboards. These dashboards display the local build under realistic load, showing new metric or log labels right when a test ran. For engineering teams using agentic coding pipelines, the tools mainly add a verification layer. They link agent output to actual system behaviour. This shifts the focus from \"did the agent understand the ticket?\" to \"does the running system match our expectations?\" This distinction matters, especially since review agents help approve changes without careful reading.", "url": "https://wpnews.pro/news/grafana-s-gcx-and-mcp-server-reach-ga-for-telemetry-driven-agent-development", "canonical_source": "https://www.infoq.com/news/2026/08/grafana-mcp-server-telemetry/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=global", "published_at": "2026-08-17 10:00:00+00:00", "updated_at": "2026-08-17 10:13:11.657377+00:00", "lang": "en", "topics": ["ai-tools", "developer-tools", "artificial-intelligence"], "entities": ["Grafana Labs", "gcx", "Grafana MCP server", "Grafana Cloud", "OpenTelemetry", "Tempo", "k6", "Claude Code"], "alternates": {"html": "https://wpnews.pro/news/grafana-s-gcx-and-mcp-server-reach-ga-for-telemetry-driven-agent-development", "markdown": "https://wpnews.pro/news/grafana-s-gcx-and-mcp-server-reach-ga-for-telemetry-driven-agent-development.md", "text": "https://wpnews.pro/news/grafana-s-gcx-and-mcp-server-reach-ga-for-telemetry-driven-agent-development.txt", "jsonld": "https://wpnews.pro/news/grafana-s-gcx-and-mcp-server-reach-ga-for-telemetry-driven-agent-development.jsonld"}}