Faultline

io.fltln由 fltln.io 发布官方发布者v2.0.0更新于 Nov 3, 2026

Check infrastructure health, manage incidents, and run runbooks in Faultline.

已验证Streamable HTTP可网页运行云端与基础设施安全与监控

概览

AI 生成的概览

Faultline 的托管 MCP 服务器，让助手检查基础设施健康状态、处理事件、查询值班人员并执行已批准的 runbook。

可以做什么
  • 列出服务清单，并查看单个服务及其最近的检查记录
  • 列出并查看事件，包括时间线、AI 摘要与事后复盘
  • 确认或解决事件，可附带备注
  • 按排班查询当前值班人员，并列出基于学习基线的延迟异常
  • 获取针对某个事件的建议下一步操作与候选 runbook
  • 对某个事件执行选定的 runbook
适用场景
适合已在使用 Faultline 的 DevOps/SRE 团队，让助手提供健康概览、解释服务为何变慢、进行事件分诊与处置，并回答值班相关问题，无需切换工具。
运行要求
远程 Streamable HTTP 端点，无需本地安装。需要 Faultline 账号，并在 Settings → API Keys 中创建 API 密钥，通过 X-API-Key 请求头（或 Bearer 令牌）发送。工具发现无需密钥，但每次工具调用都需要。
安装前请注意
该 API 密钥可访问你所在租户的监控与事件数据，应视为机密。run_runbook 会改动基础设施，可能重启或扩缩服务；确认与解决事件同样会改变状态，并可能显示在状态页上。README 称服务器不存储密钥、每次请求即时转发，且变更操作应等待明确确认。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Faultline，将其添加到工作区。
  2. 连接 X-API-Key 凭据。它们会被私密存储，绝不会在公开页面上展示。
  3. 为需要使用其工具的对话启用该服务。

Web executable，通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体，或 SourceWeft 对话中。智能体会读取该服务器的安装指南，先向你展示发布者、端点和登录方式，经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/io.fltln%2Ffaultline/install.md，并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前，先告诉我它的发布者、端点和登录方式，等我确认。不要把 API 密钥写进命令或我会提交的文件；问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "faultline": {
      "type": "http",
      "url": "https://mcp.fltln.io/mcp",
      "headers": {
        "X-API-Key": "<YOUR_API_KEY>"
      }
    }
  }
}

README

Faultline MCP Server

Faultline is infrastructure monitoring and incident management for DevOps/SRE teams. This repo documents faultline-mcp, Faultline's remote MCP server, which lets AI agents (Claude, Claude Code, Claude Desktop, or anything MCP-compatible) operate Faultline: check infrastructure health, inspect and act on incidents, look up who's on call, and run approved runbooks.

This repo is documentation plus an Agent Skill. The server is hosted by Faultline at https://mcp.fltln.io/mcp (Streamable HTTP transport), so there's nothing to install or run yourself.

Setup

  1. Create an API key in Faultline: Settings → API Keys. Keys look like flt_....
  2. Point your MCP client at https://mcp.fltln.io/mcp, sending the key as either X-API-Key: flt_... or Authorization: Bearer flt_....

Claude Code:

bash
claude mcp add --transport http faultline-mcp https://mcp.fltln.io/mcp \  --header "X-API-Key: flt_..."

Clients that take raw JSON config (Claude Desktop, etc.):

json
{  "mcpServers": {    "faultline": {      "type": "http",      "url": "https://mcp.fltln.io/mcp",      "headers": { "X-API-Key": "flt_..." }    }  }}

Discovery without a key

You can browse the toolset before creating a key. The MCP discovery methods initialize, notifications/initialized, and tools/list work without an API key, so connector directories, registry scanners, and curious humans can see every tool's name, description, input schema, and annotations:

bash
curl -s https://mcp.fltln.io/mcp \  -H "Content-Type: application/json" \  -H "Accept: application/json, text/event-stream" \  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Anything else, including every tools/call, needs a key and returns 401 without one.

Agent Skill

skills/faultline/SKILL.md is an Agent Skill that teaches Claude how to use these tools well:

  • Workflows: which tools to call for a health overview, "why is X slow?", incident triage and remediation, and on-call lookups.
  • Approval rules: read-only tools run freely; acknowledging, resolving, and running runbooks wait for your explicit go-ahead, and Claude only runs a runbook that diagnose_incident returned for that incident.

Install it alongside the MCP server:

  • Claude Code: copy the skills/faultline folder into ~/.claude/skills/ (all projects) or .claude/skills/ (one project).
  • Claude.ai / Claude Desktop: zip the faultline folder and upload it under Settings → Capabilities → Skills.

Tools

ToolWhat it does
list_servicesMonitor inventory with current status (optional status filter)
get_serviceOne service + its 10 most recent checks (for diagnosis)
list_incidentsOpen incidents (or status: "resolved" for history)
get_incidentFull incident record: timeline, AI summary, post-mortem
acknowledge_incidentAcknowledge an incident, which stops further escalation
resolve_incidentResolve with an optional note (recorded on the timeline and visible on your status page)
who_is_on_callCurrent on-call per schedule, with shift end time
list_anomaliesRecent learned-baseline latency anomalies (observed vs baseline, z-score, hours sustained, auto-opened incident if any)
diagnose_incidentRecommend the next action (run runbook / escalate / resolve / wait) + candidate runbooks. Analysis only; changes nothing
run_runbookExecute one chosen runbook against an incident. Mutates infrastructure (can restart/scale services)

Security

  • Scoped to your API key. The server never stores your key. It's used only for the duration of each request, proxied straight through to Faultline's API.
  • Keyless means read-only metadata. Requests without a key can only list the static tool definitions. They never reach Faultline's API and can't read or change any tenant data.
  • Approval-gated mutation. run_runbook is the only tool that changes infrastructure. Its description, and the Agent Skill, instruct the calling agent to use it only after diagnose_incident recommended it and you've explicitly confirmed.
  • Tenant-isolated. Every request is scoped to the tenant that owns the API key, so one key can never see or affect another tenant's data.

Support

Questions or issues: [email protected] or the Faultline dashboard.

[MCP Badge]

来源：README.md，提交 e4eb376

工具

10

list_services

List Services

List all monitored services with their current status (OK, DEGRADED, DOWN, PAUSED, PENDING). Call this first to get an overview of infrastructure health, or to find a service's ID.

输入参数status

get_service

Get Service

Get one service's details including its 10 most recent health checks (status codes, latency, errors). Call this when diagnosing WHY a service is degraded or down.

输入参数serviceId必填

list_incidents

List Incidents

List incidents — open ones by default, or recently resolved with status='resolved'. Call this when asked about outages, what's broken, or incident history.

输入参数status

get_incident

Get Incident

Get an incident's full record: timeline of every update, escalations, runbook results, AI summary, and post-mortem if drafted. Call this before acting on an incident.

输入参数incidentId必填

acknowledge_incident

Acknowledge Incident

Acknowledge an open incident, stopping further escalation. Call this when the user says they are looking into an incident or asks you to take ownership of it.

输入参数incidentId必填

resolve_incident

Resolve Incident

Mark an incident as resolved, with an optional note describing the fix. Only call this when the user confirms the underlying issue is fixed — resolving is a status-page-visible action.

输入参数noteincidentId必填

who_is_on_call

Who Is On Call

Get the person currently on call for each on-call schedule, and when their shift ends. Call this when deciding who to contact about an incident.

无需输入参数。

list_anomalies

List Anomalies

List recent learned-baseline latency anomalies: observed vs baseline response time, the deviation z-score, how many consecutive hours it has been sustained, and the incident it auto-opened (if any). Call this when asked why a service is slow, about latency spikes, or for early degradation signals that have not yet tripped a hard up/down threshold.

无需输入参数。

diagnose_incident

Diagnose Incident

Analyze an open incident and recommend the next action: run a specific runbook, escalate to a human, resolve, or wait. Returns the recommendation (with rationale and confidence) and the candidate runbooks. This is analysis ONLY — it does not change anything. Call this to decide what to do about an incident; then confirm with the user before calling run_runbook.

输入参数incidentId必填

run_runbook

Run Runbook

Execute one specific runbook against an incident. This MUTATES INFRASTRUCTURE (it can restart or scale services). Only call after diagnose_incident recommended it AND the user has confirmed. Pass the runbookId from the diagnosis's recommendation or candidate list.

输入参数runbookId必填incidentId必填

版本历史

1
  1. v2.0.0最新Sep 16, 2026

评分

已满足适用检查项的 90%

该评分衡量此收录条目的完整程度，依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价，也不表示服务器安全或当前可用：可连通性是带日期的检查结果，评分也不包含安装量或用户评价。

  • 近 30 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 已列出工具必需15 分
  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分

    请在仓库或注册中心条目中声明许可证。