Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
作者提供的说明(英文原文)
Prometheus MCP server for AI agents: PromQL queries, alert explanations, series discovery.
这三项永远不会合并成一个分数:意见、测量和人气是不同的东西。
我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。
选择你使用的 AI 应用,然后复制设置。
PROM_URL, PROM_TOKEN, PROM_TIMEOUT标为机密的值是你账户的钥匙。千万不要分享给别人,也不要粘贴到聊天里。
你用的是哪个应用?
在上方选择你的应用。我们会记住它,下次按钮会显示“安装到 …”。
链接指向作者自己的文件;我们不托管也不重新打包它。打开文件后,Claude Desktop 会进行安装。
在终端中运行这条命令
运行前请先填写空着的值。
浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
这个服务器在你的电脑上运行,而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。
你的选择通过 Cookie 保存在这台设备上,无需账户。
PROM_URL envPROM_TIMEOUT env这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。
它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月7日
注册名称已通过 GitHub 账户 aniketatgithub 验证,该账户拥有此仓库
我们还没有从 GitHub 读取这个仓库
我们还没有从 GitHub 读取这个仓库
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
它在你的电脑上运行;我们只能读取互联网端点的工具列表
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
你需要填写:PROM_TOKEN
它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。
GitHub 仓库最后更新于–,最新版本(0.2.0)于28天前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 MCPB。
同一分类中有:Spec Driven Development, Remote Desktop Commander, FactLot
同一分类中其他作者的服务器,按星标数排序。
如何看卡片:★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Cited CRE facts and deterministic calculators for agents: rent, CAM, debt sizing, IRR, parcel IDs.
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
prom-mcp is a Prometheus MCP server (Model Context Protocol) that gives AI agents production-grade eyes on your metrics. Connect Claude Code, Cursor, or any MCP client to Prometheus and let the agent run PromQL queries, explain firing alerts with their rule definitions, and discover which series actually exist before querying — so it investigates instead of guessing.
Built by a production engineer who works on-call, so the output reads like a triage note, not a raw JSON dump.
5 tools • 2 resources • 1 triage prompt • single Go binary • zero dependencies • stdio
Download the bundle for your platform, then open it or drag it into Claude Desktop (Settings → Extensions). You will be asked for your Prometheus URL, which defaults to http://localhost:9090.
| Platform | MCPB bundle |
|---|---|
| macOS (Apple silicon) | prom-mcp_v0.2.0_darwin_arm64.mcpb |
| macOS (Intel) | prom-mcp_v0.2.0_darwin_amd64.mcpb |
| Linux (x86_64) | prom-mcp_v0.2.0_linux_amd64.mcpb |
| Linux (arm64) | prom-mcp_v0.2.0_linux_arm64.mcpb |
With the binary on your PATH (release download or go install, see below):
claude mcp add prom -- prom-mcp
Set PROM_URL in your shell first if your Prometheus is not at http://localhost:9090, or pass --env PROM_URL=... to the command. More variants in examples/claude-code.md.
Add to ~/.cursor/mcp.json (same file as examples/cursor-mcp.json):
{
"mcpServers": { "prom": { "command": "prom-mcp", "env": { "PROM_URL": "http://localhost:9090" } } } }
Add to .vscode/mcp.json (same file as examples/vscode-mcp.json):
{
"servers": { "prom": { "command": "prom-mcp", "env": { "PROM_URL": "http://localhost:9090" } } } }
go install github.com/aniketatgithub/prom-mcp@latest
git clone https://github.com/aniketatgithub/prom-mcp && cd prom-mcp
go build -o prom-mcp ./cmd/prom-mcp
Plain binaries for all four platforms are also on the v0.2.0 release, next to the MCPB bundles above.
| You ask | prom-mcp does |
|---|---|
| "Why is NodeDown firing?" | prom_alerts_explain returns the active alert joined with its rule expression, group, for, labels, and annotations. |
| "What jobs and instances actually exist?" | prom_series_discover and prom_label_values list real series and label values before any query is written. |
| "Is this getting worse?" | prom_query_range summarizes the window per series: points, first, last, min, max. |
Alert triage: why is the node down?
Discover what exists before querying:
Check the trend: is node1 getting worse?
Ask your agent *"why is the node down?"* and the alert tool answers in triage shape:
1 active alert(s)
[firing] NodeDown (since 2026-10-08T10:00:00Z, value 0) labels: {alertname="NodeDown", instance="node1:9100", job="node", severity="critical"} summary: Node node1:9100 is down description: Scrapes failing for 5m rule: up{job="node"} == 0 group: node.rules for: 5m0s
Plus a built-in triage prompt template that walks the agent through alerts, discovery, and trend checks in order.
| Tool | What it does |
|---|---|
prom_query | Instant PromQL query, rendered as compact labelled series and values. |
prom_query_range | Range query over the last N minutes; per-series points, first, last, min, max. |
prom_alerts_explain | Active alerts joined with alerting-rule expressions, rule group, for duration, and annotations. |
prom_label_values | Lists real values of a label (optionally scoped), so the agent learns actual job/instance names. |
prom_series_discover | Checks which series exist for a label matcher, so the agent stops hallucinating metric names. |
Resources: prometheus://alerts (explained alerts), prometheus://config. Prompt: triage.
prom-mcp demo # fixture-backed self-test: query, alert explanation, series discovery
prom-mcp query 'up' # one-shot CLI query against $PROM_URL
| Setting | Env var | Config file key |
|---|---|---|
| Prometheus base URL | PROM_URL | base_url |
| Bearer token (auth-protected Prometheus) | PROM_TOKEN | token |
| Request timeout | PROM_TIMEOUT (e.g. 30s) | timeout_seconds |
| Config file path | PROM_CONFIG |
Config file is JSON, looked up at $PROM_CONFIG, then ./prom-mcp.json, then ~/.prom-mcp.json:
{ "base_url": "https://prometheus.example.com", "token": "…", "timeout_seconds": 30 }
Env vars override the file. No config at all defaults to http://localhost:9090.
Agents drown in raw Prometheus JSON and invent metric names that do not exist. prom-mcp returns terse, labelled, triage-shaped text and makes the agent discover before querying. That is the difference between an agent that guesses and an agent that investigates.
Single Go binary, zero dependencies, works fully offline against your own Prometheus.
Comparing servers? See docs/comparison.md for an honest side-by-side with pab1it0/prometheus-mcp-server and the Prometheus org's prometheus/prometheus-mcp.
Which AI clients work with prom-mcp? Any MCP client: Claude Code, Cursor, Windsurf, Zed, and custom agents. It speaks newline-delimited JSON-RPC 2.0 over stdio.
How do I connect prom-mcp to a remote Prometheus? Set PROM_URL to your Prometheus base URL (for example http://localhost:9090 or your internal endpoint) when adding the server.
Can an AI agent explain why a Prometheus alert is firing? Yes. prom_alerts_explain joins active alerts with the rule that fired them (expression, summary, description), so the agent can reason about the cause instead of reporting raw JSON.
Does prom-mcp need any API keys or cloud services? No. It is a single local binary talking only to your Prometheus.
How does prom-mcp compare to the other Prometheus MCP servers? See docs/comparison.md. Short version: prom-mcp is the focused alert-triage option with 5 tools; the others cover more of the Prometheus API or ship more deployment machinery.
How is this different from exposing the Prometheus HTTP API to an agent? Raw API access returns nested JSON the agent must parse and invites invented metric names. prom-mcp renders compact labelled text, joins alerts with the rules that fired them, caps long outputs, and nudges discovery before querying. It also retries transient failures and supports bearer-token auth.
Does it work with a remote or auth-protected Prometheus? Yes: set PROM_URL and PROM_TOKEN (or a JSON config file). See Configuration.
Which Prometheus versions are supported? Any server exposing the standard HTTP API (/api/v1/query, query_range, alerts, rules, series, label/ /values).
MIT. See LICENSE.