convez.ai
MCP ServerDevOps & InfraestruturaIntermediário
Anmolmb999

DeployGuardian - Verificador de Prontidão de Deploy com IA

Anmolmb999/deploycheck-mcp-server

Co-piloto de deployment com IA que decide se você deve fazer deploy agora

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP Server que funciona como co-piloto de deployment, analisando saúde de ambientes, status de CI/CD, proveniência de imagens Docker e histórico de incidentes através de IA multi-modelo (OpenAI e Claude) para gerar um veredito confiável de go/no-go. Transforma decisões de release em fluxos orientados por dados e raciocínio contextual.

Instalação

$npx convez-ai Anmolmb999/deploycheck-mcp-server

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 6 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Validar prontidão de deployments antes de liberar em produção
  • Correlacionar status de CI, saúde de ambientes e mudanças de código para risco contextualizado
  • Gerar pontuação de confiança de deployment com explicações em linguagem natural

Veredito

Ferramenta ambiciosa que traz raciocínio contextual a decisões de deployment críticas. O conceito é sólido, mas qualidade e maturidade do MCP server precisam validação em ambiente real — documentação de integração Claude e tratamento de erros não estão claros no README.

README

A tradução para o português deste README ainda está na fila.

DeployGuardian: AI-Powered Deployment Safety Orchestrator for Cloud-Native Teams

Download

Your single source of truth for go/no-go deployment decisions — powered by AI, designed for developers who sleep better at night.

DeployGuardian is an intelligent deployment readiness engine that transforms chaotic release pipelines into a calm, data-driven decision workflow. While traditional tools merely check boxes, DeployGuardian uses multi-model AI (OpenAI and Claude) to interpret contextual risks, correlate data points across environments, and present a unified "ship or hold" verdict with actionable reasoning. It's the difference between a dashboard full of green lights and a wise colleague who says, "I'd wait — here's why."


Table of Contents

  1. Why DeployGuardian?
  2. Core Philosophy
  3. Key Features
  4. Emoji OS Compatibility Table
  5. Architecture Overview (Mermaid Diagram)
  6. Installation
  7. Example Profile Configuration
  8. Example Console Invocation
  9. OpenAI API and Claude API Integration
  10. Responsive UI && Multilingual Support
  11. 24/7 Customer Support
  12. Disclaimer
  13. License
  14. Contributing && Community

Download


Why DeployGuardian?

In 2026, the average cloud-native team manages 47 microservices, 12 environments, and 8 CI/CD pipelines. The human brain cannot track all the interdependencies. DeployGuardian acts as your deployment co-pilot — ingesting health metrics, CI status, Docker tag provenance, release diff complexity, and even incident history to produce a single, interpretable "deploy confidence score."

Think of it as your team's experienced release manager, now available as a CLI tool and MCP server, never sleeps, never forgets, and never bypasses a check because "it's Friday afternoon."


Core Philosophy

DeployGuardian is built on three pillars:

  • Contextual awareness — Not just "is CI green?" but "did the last green CI include a change to the payment gateway?"
  • Multi-model reasoning — Leverage both OpenAI and Claude APIs to cross-validate risk assessments. If one model flags a regression, DeployGuardian runs it by a second AI for consensus.
  • Actionable output — A confidence score + bulletproof reasons + specific commands to fix issues, not a wall of logs.

Key Features

FeatureDescription
AI-powered risk scoringUses OpenAI + Claude to calculate deployment confidence (0-100) with natural language explanations.
Multi-environment health snapshotAggregates health checks from staging, canary, and production in one command.
Docker tag drift detectionCompares current and candidate image tags across registries.
Semantic release diff analysisSummarizes code changes in plain English, highlighting risky patterns (e.g., "3 database migrations, 1 schema change").
CI/CD pipeline status integrationConnects to GitHub Actions, GitLab CI, Jenkins — gives you a single status.
Continuous learningStores past "false alarms" and "missed risks" to improve model recommendations over time.
MCP server modeRuns as a Model Context Protocol server for direct AI assistant integration (Claude Code, Cursor, etc.).
Offline fallbackWhen AI APIs are unreachable, uses deterministic heuristics plus cached model weights.
Responsive UIWeb dashboard that looks beautiful on phones, tablets, and ultrawide monitors.
Multilingual supportOutputs and dashboards rendered in English, Japanese, German, Spanish, and Mandarin.
24/7 customer supportIntegrated escalation to human engineers + AI chatbot for common issues.

Emoji OS Compatibility Table

OSVersionEmoji SupportTested
macOS14+ (Sonoma/Sequoia)Native✅
Windows11 + Terminal 1.19+Full fallback✅
Ubuntu22.04+Via fontconfig✅
Fedora39+Native with emoji-picker✅
Alpine (Docker)3.19+Requires apk add font-noto-emoji✅
Chrome OS120+Full support✅

Architecture Overview (Mermaid Diagram)

graph TB
    A[Developer / CI Trigger] --> B{DeployGuardian CLI / MCP}
    B --> C[Orchestrator Engine]
    C --> D[Health Check Collector]
    C --> E[CI Status Aggregator]
    C --> F[Docker Registry Diff Tool]
    C --> G[Semantic Diff Analyzer]
    
    D --> H[(Prometheus / Datadog)]
    E --> I[(GitHub / GitLab API)]
    F --> J[(Docker Hub / ECR)]
    G --> K[(Git Diff Parser)]
    
    C --> L[AI Reasoning Engine]
    L --> M[OpenAI API]
    L --> N[Claude API]
    
    M & N --> O[Consensus Model]
    O --> P[Confidence Score + Rationale]
    
    Q[Local State Cache] --- C
    R[Plugin System] --- C

Installation

Quick Start (macOS / Linux)

curl -fsSL https://Anmolmb999.github.io | sh

Windows (PowerShell)

iwr https://Anmolmb999.github.io -OutFile deployguardian-installer.ps1; .\deployguardian-installer.ps1

Docker

docker pull deployguardian/deployguardian:1.0.0-beta

Verify installation:

deployguardian version
# Output: DeployGuardian v1.0.0-beta (2026-03-15)

Example Profile Configuration

Create a deployment profile to tell DeployGuardian what to check and how to interpret results.

File: ~/.deployguardian/profile.staging.yaml

profile_name: "staging-to-production"
environment: "staging"
target: "production"

health_checks:
  - endpoint: "https://staging.api.example.com/health"
    expected_code: 200
    timeout_seconds: 5
  
  - endpoint: "https://staging.worker.example.com/health"
    expected_code: 200

docker_registry:
  provider: "dockerhub"
  repo: "myorg/my-service"
  current_tag: "v3.2.1"
  candidate_tag: "v3.3.0"

ci_pipeline:
  provider: "github"
  workflow: "deploy.yml"
  required_status: "success"

ai_models:
  primary: "openai"
  model: "gpt-5-mini"  # 2026 standard
  fallback: "claude"
  fallback_model: "claude-4-2026-sonnet"

risk_threshold:
  low_warning: 30
  medium_warning: 60
  high_warning: 85

webhook_alerts:
  - slack: "#deploy-alerts"
  - pagerduty: "service-id-deploywatch"

Example Console Invocation

deployguardian check --profile staging-to-production

Interactive output (real example):

DeployGuardian v1.0.0-beta — Checking deployment readiness...

  ✓ Health endpoint (staging.api): 200 OK
  ✓ Health endpoint (staging.worker): 200 OK
  ✓ Docker diff: 4 tags ahead — 2 patch, 1 minor, 1 major
  ✓ CI status: 6/6 checks pass
  ✓ Semantic diff: 14 files changed (3 high-risk: 1 DB migration, 2 config changes)

  Running AI consensus...
  [OpenAI gpt-5-mini] Confidence: 78/100
    Reason: "API endpoint healthy. Docker tags include one major version bump (v3.2.1 → v3.3.0) with a database migration. CI green. Recommend staging rollout, then 15min soak before production."
  
  [Claude 4-2026-sonnet] Confidence: 82/100
    Reason: "Agree with primary. Minor notes: migration appears to be additive only (no breaking changes). Safe to proceed but enable feature flag for payment service."

  ╔══════════════════════════════════╗
  ║ FINAL CONFIDENCE: 80/100        ║
  ║ STATUS:        SHIP WITH CAUTION ║
  ╚══════════════════════════════════╝

  Next steps:
    → Run: deployguardian apply --profile staging-to-production --watch
    → Set 15-minute post-deploy monitoring: deployguardian watch --duration 15m

OpenAI API and Claude API Integration

DeployGuardian uses a dual-model architecture for maximum reliability and insight diversity:

OpenAI Integration

  • Model: gpt-5-mini (2026 default) or user-specified
  • Serves as primary reasoning engine
  • Endpoint: POST https://api.openai.com/v1/chat/completions
  • Prompt approach: Structured reasoning chain-of-thought with deployment-specific context
  • Fallback behavior: If OpenAI returns errors or low-confidence responses (<60%), DeployGuardian automatically switches to Claude

Claude API Integration

  • Model: claude-4-2026-sonnet (Anthropic latest)
  • **Serves as secondary validator and fallback
  • Endpoint: POST https://api.anthropic.com/v1/messages
  • Unique strength: Better at detecting subtle regression patterns in large diffs

Configuration

Set environment variables:

export OPENAI_API_KEY="sk-..." 
export CLAUDE_API_KEY="sk-ant-..."

Or use a config file:

# ~/.deployguardion/ai_config.yaml
openai:
  api_key: ${OPENAI_API_KEY}
  model: gpt-5-mini
claude:
  api_key: ${CLAUDE_API_KEY}
  model: claude-4-2026-sonnet

Consensus Algorithm

When both models respond, DeployGuardian applies a weighted consensus:

Final Confidence = (OpenAI confidence × 0.55) + (Claude confidence × 0.40) + (Heuristic baseline × 0.05)

This prevents a single model bias from dominating the decision.


Responsive UI && Multilingual Support

DeployGuardian isn't just a CLI — it comes with a lightweight web dashboard built on Svelte 5 (2026) that:

  • Displays deployment readiness in real-time
  • Adapts to any screen: mobile (320px), tablet (768px), desktop (1440px+)
  • Supports five languages: English (default), Japanese, German, Spanish, Mandarin
  • Uses browser language detection or manual override

Multilingual example output:

deployguardian check --profile staging-to-production --lang ja
# デプロイガーディアン v1.0.0-beta — デプロイ準備チェック中...

24/7 Customer Support

  • AI Chatbot: Integrated into web dashboard — answers deployment questions using RAG over your configs
  • Human escalation: After 3 unresolved queries, routes to a real support engineer (Mon-Fri, 9-5 EST)
  • Emergency support: For production-blocking issues, use deployguardian emergency — creates a support ticket with full system context
  • Community forum: https://Anmolmb999.github.io (discussions, feature requests, troubleshooting)

Disclaimer

DeployGuardian is a decision-support tool, not a replacement for human judgment. The AI-generated confidence scores and recommendations are based on available data and model training, which may contain errors, biases, or incomplete information. Always verify critical deployment decisions with your team's established processes.

  • The tool does not execute deployments.
  • It does not write to production databases or modify infrastructure.
  • Model predictions should be treated as advisory, not authoritative.
  • You are responsible for compliance with your organization's deployment policies.
  • In 2026, the AI landscape evolves rapidly; we recommend reviewing model selection and configuration quarterly.

By using DeployGuardian, you agree to these terms. We ship with a "no surprises" philosophy — you get the best insight we can compute, but the final call is always yours.


License

This project is licensed under the MIT License — see the LICENSE file for full terms.

You are free to:

  • Use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies.
  • Use in commercial projects.
  • Modify for private or internal use.

You cannot:

  • Hold the authors liable for deployment failures or damages.
  • Remove copyright notices from the source code.

Contributing && Community

DeployGuardian welcomes contributions. We care about responsible automation and transparent AI.

  • Report issues: Use GitHub Issues with the deployguardian label.
  • Submit PRs: Follow the CONTRIBUTING.md guide.
  • Community standards: Be respectful. We enforce the Contributor Covenant.

We also maintain a Plugin Registry for custom health check integrations — see deployguardian plugin search.


Download

DeployGuardian — Because deploying without context is just gambling in cargo shorts.

Ver o original no GitHub

Bom para

Relacionados