groundcover 让编程智能体接入生产可观测:MCP 拉证据或异步交办根因分析
可观测平台 groundcover 为 Claude、Codex、Cursor 与 AWS Kiro 等编程智能体提供生产侧证据:既可经 MCP 拉取日志/链路/K8s 事件,也可把模糊问题异步交给其观测智能体做根因分析。
编程智能体很擅长根据仓库推断「代码应当怎样表现」,但一过部署边界就常常失明。可观测平台 groundcover 补的正是这一段:让 Claude、Codex 等把生产排查交给自家观测智能体,Cursor 与 AWS Kiro 则可通过 MCP 工具接入。
接入分两条路。若智能体已经知道要看什么证据,可直接经 MCP 拉取日志、链路与 Kubernetes 事件;若问题更含糊,则作为一个异步 Agent Task 交给 groundcover Agent——立刻返回任务 ID,调用方继续干别的,稍后取回根因分析或草稿仪表盘。产品内还有 Agent Mode,工程师可实时纠偏调查方向,或让智能体根据发现生成监控与仪表盘;在 Slack 事故线程里 @groundcover,会把对话当作上下文,并可联动 Linear、GitHub 等已连接系统。
架构上它走客户云内部署(bring-your-own-cloud),遥测留在客户自己的云账号,计费按节点而非按数据量——这对需要保留高保真证据的智能体排查更友好。联合创始人兼 CEO Shahar Azulay 的表述很直接:没有证据,智能体就说不清生产里「已经发生了什么」。编辑器里的智能体正在往生产侧延伸;会读代码,也开始能读线上事实。
groundcover gives coding agents a way to see production
groundcover just gave coding agents a way to see production.
Claude and Codex can now hand production investigations to groundcover's own observability agent. Cursor and AWS's Kiro get in through MCP tools.
Coding agents can review a pull request and work out from the code how software is meant to behave, but their view usually ends at deployment. That is the gap it closes. groundcover co-founder and CEO Shahar Azulay put it directly: agents "can't tell you what already did happen in production unless they have access to the evidence."
Two ways in. An agent that already knows which evidence it wants pulls logs, traces, and Kubernetes events straight from groundcover's MCP tools. A vaguer problem goes to the groundcover Agent as an asynchronous Agent Task. A job ID comes back right away, the investigation runs in the background while the calling agent moves on, and the findings arrive later as a completed root cause analysis or a draft dashboard.
Inside groundcover itself, Agent Mode lets engineers steer an investigation in real time or have the agent generate dashboards and monitors from what it finds. In Slack, teams can bring @groundcover into an incident thread, where it treats the conversation as context and can draw on connected systems like Linear and GitHub.
The architecture choice matters too. It runs on groundcover's bring-your-own-cloud setup, so telemetry stays inside the customer's own cloud account, and customers pay by the node instead of by data volume. Keeping more evidence at higher fidelity doesn't inflate the bill, and an agent needs that depth of evidence to investigate accurately.
The pattern is clear now. Agents started in the editor, and they're moving into production. The coding agent knows what the code should do. Now it can see what it did.