MCP Health Monitor

io.github.oaslananka由 GitHub 账号 oaslananka 发布v1.4.0更新于 Oct 24, 2026

Monitor MCP, CI, and HTTP targets with health history, TLS expiry, assertions, and diagnostics

概览

AI 生成的概览

本地 MCP 服务器,让助手登记并巡检 MCP 服务器、GitHub Actions、GitLab 流水线与 HTTP 端点,并保留历史与报告。

可以做什么
登记、列出和移除监控目标:MCP 服务器(Streamable HTTP、SSE、需显式开启的 stdio)、GitHub Actions 工作流、GitLab 流水线和仅 GET 的 HTTP 端点,以及 webhook 告警目标和测试发送。对单个或全部目标执行实时检查,支持状态码、响应头、正文、JSON 与 TLS 到期断言,并给出失败作业或步骤诊断。查看历史与报告:可用率与延迟、跨提供方的 JSON 仪表盘、Markdown 报告和监控统计。配置 MCP 告警阈值,如响应时间、可用率百分比和连续失败次数。
适用场景
适合让助手持续关注 CI 流水线、MCP 端点或服务健康端点,并按时间汇总失败、延迟或证书到期情况。适合希望用一个本地工具按需或定时检查并生成 Markdown 报告的运维和开发人员。
运行要求
通过 npx 以 stdio 在本地运行,需要 Node.js 24。可选的 GITHUB_TOKEN 和 GITLAB_TOKEN 用于私有仓库和更高速率限制,仅保存变量名。自建 GitLab 和私有 HTTP 来源需通过环境变量加入允许列表。本地 stdio 检查需设置 HEALTH_MONITOR_ALLOW_STDIO 并提供可执行文件允许列表。Webhook 密钥需要 HEALTH_MONITOR_WEBHOOK_ENCRYPTION_KEY。历史保存在本地 SQLite 数据库。
安装前请注意
登记 webhook 会保存密钥,未设置 HEALTH_MONITOR_WEBHOOK_ENCRYPTION_KEY 时会被拒绝;输出中密钥会被隐去。GITHUB_TOKEN 和 GITLAB_TOKEN 在检查时读取且不落盘,但可读取私有仓库和流水线。登记与注销目标会写入本地 SQLite 数据库,注销会删除该目标的历史。开启 stdio 检查会运行本地可执行文件,应保持允许列表尽量小。测试 webhook 会向外部地址发送请求。

安装

在 SourceWeft 中

  1. 打开 控制台中的 MCP Health Monitor,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

health-monitor-mcp

MCP server, CI workflow, and generic HTTP endpoint monitoring with health history, TLS expiry, assertions, diagnostics, and operational reports.

[CI] [codecov] [Release] [License: MIT] [OpenSSF Scorecard] [Node.js]

What This Does

health-monitor-mcp keeps local registries of MCP servers, GitHub Actions workflows, GitLab pipelines, and generic HTTP endpoints. It performs live checks, records bounded evidence in SQLite, evaluates MCP alert thresholds, and returns JSON or Markdown diagnostics suitable for agents and operators.

Supported target transports:

  • Streamable HTTP for current remote MCP servers.
  • SSE for legacy MCP servers.
  • stdio for trusted local executables after explicit opt-in.
  • GitHub Actions workflow runs, failed jobs, and failed steps for public or private repositories.
  • GitLab CI/CD pipelines, failed jobs, stages, refs, commits, URLs, and bounded trace excerpts for GitLab.com or allowlisted self-hosted instances.
  • Generic HTTP/HTTPS endpoints with status, header, body-substring, JSON-value, redirect, latency, and TLS-expiry checks.

Azure DevOps monitoring was retired in v1.1.0. GitHub Actions shipped in v1.2.0 and GitLab CI/CD in v1.3.0; generic HTTP/TLS monitoring completes the current multi-provider feature line.

Quick Start

Run the published package noninteractively with Node.js 24:

bash
npx -y health-monitor-mcp --version

Example MCP client configuration:

json
{  "mcpServers": {    "health-monitor": {      "command": "npx",      "args": ["-y", "health-monitor-mcp"]    }  }}

Tools

ToolPurposeTypical prompt
register_serverRegister an MCP targetRegister inventory-prod
check_serverRun one MCP health checkCheck inventory-prod now
register_github_actionsRegister a GitHub workflowMonitor ci.yml in owner/repo
check_github_actionsCheck latest GitHub run and failed job/step diagnosticsCheck repo-ci now
register_gitlab_pipelineRegister a GitLab project pipelineMonitor group/project on GitLab
check_gitlab_pipelineCheck latest pipeline and bounded failed-job tracesCheck gitlab-ci now
register_http_targetRegister a GET-only HTTP/HTTPS endpointMonitor the public health endpoint
check_http_targetCheck response assertions and TLS expiryCheck service-health now
register_webhookRegister a webhook target for alert deliveryRegister alert-webhook for notifications
test_webhookSend a test payload and return delivery diagnosticsTest alert-webhook delivery
check_allCheck all target kinds with bounded concurrencyCheck all production targets
get_uptimeReturn MCP uptime and latency historyShow 24h uptime for inventory-prod
get_dashboardReturn a cross-provider JSON dashboardGive me a 24h dashboard
get_reportReturn a cross-provider Markdown reportGenerate a 24h health report
list_serversList registered MCP targetsList monitored MCP servers
list_github_actionsList registered GitHub workflow targetsList monitored workflows
list_gitlab_pipelinesList registered GitLab pipeline targetsList monitored GitLab pipelines
list_http_targetsList registered HTTP targetsList monitored HTTP endpoints
list_webhooksList registered webhook targetsList webhook targets
unregister_serverRemove an MCP targetStop monitoring local-debugger
unregister_github_actionsRemove a GitHub target and its historyStop monitoring repo-ci
unregister_gitlab_pipelineRemove a GitLab target and its historyStop monitoring gitlab-ci
unregister_http_targetRemove an HTTP target and its historyStop monitoring service-health
unregister_webhookRemove a webhook target and its historyStop monitoring alert-webhook
set_alertConfigure MCP health thresholdsAlert if inventory-prod exceeds 500ms
get_monitor_statsInspect cross-provider monitor activityHow many checks are stored?

Expected configuration mistakes return stable error codes and remediation hints, including SERVER_NOT_FOUND, GITHUB_ACTIONS_TARGET_NOT_FOUND, GITLAB_PIPELINE_TARGET_NOT_FOUND, GITLAB_BASE_URL_NOT_ALLOWED, HTTP_TARGET_NOT_FOUND, HTTP_TARGET_URL_NOT_ALLOWED, NO_SERVERS_REGISTERED, STDIO_DISABLED, STDIO_COMMAND_REJECTED, WEBHOOK_ENCRYPTION_KEY_REQUIRED, WEBHOOK_REGISTRATION_FAILED, and WEBHOOK_NOT_FOUND.

Register Targets

Streamable HTTP:

text
register_server name="inventory-prod" type="http" url="https://inventory.example.com/mcp" tags=["production","inventory"]

Legacy SSE:

text
register_server name="legacy-search" type="sse" url="https://search.example.com/sse" tags=["legacy"]

Trusted local stdio:

bash
export HEALTH_MONITOR_ALLOW_STDIO=1export HEALTH_MONITOR_STDIO_ALLOWLIST=npx,node
text
register_server name="local-debugger" type="stdio" command="npx" args=["-y","mcp-debug-recorder"] tags=["local"]

The command field must contain one executable only. Put package names and flags in args. Remote-safe runtime profiles always disable stdio.

Register GitHub Actions

Public repositories can be checked without authentication. Private repositories and higher API rate limits require a token with Actions read access:

bash
export GITHUB_TOKEN=your-runtime-secret
text
register_github_actions name="repo-ci" owner="oaslananka" repository="health-monitor-mcp" workflow="ci.yml" branch="main" token_env="GITHUB_TOKEN" tags=["production","ci"]check_github_actions name="repo-ci" timeout_ms=5000

Only the environment-variable name in token_env is stored. The token value is read at check time and is never written to SQLite, logs, reports, or tool responses.

Register GitLab Pipelines

Public GitLab.com projects can be checked without authentication. Private projects require a token exposed only through the runtime environment:

bash
export GITLAB_TOKEN=your-runtime-secret
text
register_gitlab_pipeline name="gitlab-ci" project="group/project" ref="main" token_env="GITLAB_TOKEN" tags=["production","ci"]check_gitlab_pipeline name="gitlab-ci" timeout_ms=5000

GitLab.com is allowed by default. A self-hosted instance must use an HTTPS origin and be explicitly allowed:

bash
export HEALTH_MONITOR_GITLAB_BASE_URL_ALLOWLIST=https://gitlab.internal.example
text
register_gitlab_pipeline name="private-gitlab" base_url="https://gitlab.internal.example" project="platform/service" token_env="GITLAB_TOKEN"

Only token_env is persisted. Token values, response bodies, and full traces are never stored or returned. Failed-job trace excerpts are range-requested, sanitized, and bounded.

Register HTTP Targets

Public HTTP and HTTPS endpoints are allowed by default. The provider sends GET requests only and supports bounded status, header, body-substring, JSON scalar, and TLS-expiry assertions:

text
register_http_target name="service-health" url="https://status.example.com/health" expected_statuses=[200] header_assertions=[{"name":"x-ready","equals":"yes"}] body_contains=["ready"] json_assertions=[{"path":"status","equals":"ready"}] tls_expiry_days=30 tags=["production","http"]check_http_target name="service-health" timeout_ms=5000

Private, loopback, link-local, and other non-public addresses are blocked. A trusted private origin is available only in the full runtime profile and must be explicitly listed:

bash
export HEALTH_MONITOR_HTTP_TARGET_ALLOWLIST=https://status.internal.example:8443

Every DNS answer and every redirect destination is revalidated. Responses are capped at 262144 bytes; full response bodies and certificate chains are never stored or returned.

Register Webhook Targets

Webhook targets receive alert notifications via HTTPS POST with optional HMAC-SHA256 signing. Secrets are encrypted at rest when HEALTH_MONITOR_WEBHOOK_ENCRYPTION_KEY is configured (32+ characters); otherwise secret registration is rejected.

text
register_webhook name="alert-webhook" url="https://hooks.example/alerts" secret="webhook-secret" events=["down","alert"] tags=["production","alerts"]test_webhook name="alert-webhook"

Secrets are never returned in listings or diagnostics—always redacted. Delivery diagnostics include status, latency, HTTP status code, and error message.

Health Checks and Reports

text
check_server name="inventory-prod" timeout_ms=5000check_all timeout_ms=5000 tags=["production"]get_uptime name="inventory-prod" hours=24get_dashboard hours=24 include_tool_stats=trueget_report hours=24

HEALTH_MONITOR_MAX_CONCURRENCY limits MCP, GitHub Actions, GitLab, and HTTP checks through one shared scheduled and interactive queue. Results preserve MCP-then-GitHub-then-GitLab-then-HTTP registration order even when checks complete out of order.

Alerts

text
set_alert name="inventory-prod" max_response_time_ms=500 min_uptime_percent=99 consecutive_failures_before_alert=2

Alert findings are evaluated by check_server, check_all, and get_dashboard. Configure webhook targets with register_webhook to receive alert notifications.

Configuration

VariableDefaultPurpose
HEALTH_MONITOR_DB~/.mcp-health-monitor/health.dbSQLite database path
HEALTH_MONITOR_AUTO_CHECK0Enable scheduled checks with 1
HEALTH_MONITOR_RETENTION_DAYS30Health-history retention
HEALTH_MONITOR_MAX_CONCURRENCY5Scheduled and interactive check concurrency
GITHUB_TOKENunsetOptional GitHub Actions read token
GITLAB_TOKENunsetOptional GitLab project/pipeline/job read token
HEALTH_MONITOR_GITLAB_BASE_URL_ALLOWLISTunsetAllowed self-hosted GitLab HTTPS origins
HEALTH_MONITOR_HTTP_TARGET_ALLOWLISTunsetPrivate HTTP(S) origins allowed in full profile
HEALTH_MONITOR_ALLOW_STDIO0Allow trusted local stdio checks
HEALTH_MONITOR_STDIO_ALLOWLISTunsetOptional comma-separated executable allowlist
HEALTH_MONITOR_HTTP_TOKENunsetBearer token for POST /mcp
HEALTH_MONITOR_HTTP_ORIGIN_ALLOWLISTunsetAllowed remote client origins
HEALTH_MONITOR_HTTP_MAX_BODY_BYTES1048576Maximum inbound MCP body
HEALTH_MONITOR_HTTP_BODY_TIMEOUT_MS15000Inbound body read timeout
HEALTH_MONITOR_HTTP_STATEFUL_SESSIONS0Enable stateful Streamable HTTP sessions
HEALTH_MONITOR_HTTP_SESSION_TTL_MS1800000Stateful session TTL
HEALTH_MONITOR_HTTP_MAX_SESSIONS100Stateful session cap
HEALTH_MONITOR_WEBHOOK_ENCRYPTION_KEYunset32+ char key for encrypting webhook secrets
HEALTH_MONITOR_WEBHOOK_TIMEOUT_MS5000Webhook delivery timeout in milliseconds

HTTP Deployment

The server binds to 127.0.0.1 by default. A non-loopback bind requires a remote-safe profile, bearer token, and Origin allowlist.

bash
HOST=0.0.0.0 \HEALTH_MONITOR_PROFILE=remote-safe \HEALTH_MONITOR_HTTP_TOKEN=change-me \HEALTH_MONITOR_HTTP_ORIGIN_ALLOWLIST=https://client.example \npx -y health-monitor-mcp-http

GET /health is unauthenticated and exposes only status and version. POST /mcp requires Authorization: Bearer <token>.

Docker

Persist /data; otherwise the SQLite database disappears with the container.

bash
docker volume create health-monitor-data
docker run --rm \  -v health-monitor-data:/data \  -p 127.0.0.1:3000:3000 \  -e HOST=0.0.0.0 \  -e HEALTH_MONITOR_PROFILE=remote-safe \  -e HEALTH_MONITOR_HTTP_TOKEN=change-me \  -e HEALTH_MONITOR_HTTP_ORIGIN_ALLOWLIST=https://client.example \  ghcr.io/oaslananka/health-monitor-mcp:latest

Development

The repository pins Node.js 24.18.0 and pnpm 11.14.0 through .mise.toml.

bash
mise trustmise installpnpm install --frozen-lockfilepnpm run ci

Useful gates:

bash
pnpm run buildpnpm run typecheckpnpm run lintpnpm run lint:testpnpm run test:coveragepnpm run test:integrationpnpm run docs:api:checkpnpm run security:supply-chainpnpm run check:metadatapnpm run check:package

Coverage and Test Analytics

Jest remains the blocking coverage gate with repository-local thresholds. The CI validation job runs all unit and integration tests once, writes coverage/lcov.info and reports/junit/junit.xml, and uploads both reports to Codecov. Codecov project and patch statuses start as informational with target: auto and a 1% tolerance, adding pull-request diff coverage, file-level visibility, and failed-test analytics without duplicating the local merge gate.

Codecov Bundle Analysis is intentionally not enabled. This package ships Node.js entrypoints compiled with tsc; it does not currently produce a Rollup, Vite, or Webpack application bundle whose download size is a product metric.

Architecture and Roadmap

Agent Runtime Configuration

This repository owns its product-specific MCP configuration, plugin manifest, and skills:

FilePurpose
.claude-plugin/plugin.jsonClaude Code plugin manifest
.mcp.jsonProject-local MCP configuration
.codex/config.example.tomlCodex CLI example
.vscode/mcp.example.jsonVS Code / Copilot example
opencode.example.jsoncOpenCode example
skills/Product-specific monitoring workflows
docs/agent-runtime-config.mdRuntime setup and validation

Security and Contributing

Report vulnerabilities through GitHub Private Vulnerability Reporting. See SECURITY.md and docs/security.md.

Contribution setup and standards are documented in docs/contributing.md. Usage questions belong in GitHub Discussions; actionable work belongs in issues.

License

MIT

来源:README.md,提交 77e4dcc

工具

0
SourceWeft 尚未在沙箱中启动过该软件包,因此其工具未知。

版本历史

1
  1. v1.4.0最新Sep 16, 2026

评分

已满足适用检查项的 50%

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25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分