DevOps工具API服务器

Agent → custom boto3 code → AWS
Agent → custom subprocess → Terraform
Agent → custom SDK → Azure / GCP / K8s / Datadog / ...

Agent → POST /tools/execute → MCP Server → any of 20 platforms

┌────────────────────────────────────────────────────────────────────┐
│                           AI Agent                                 │
│           (Claude / LangGraph / AutoGen / custom)                  │
└────────────────────────┬───────────────────────────────────────────┘
                         │  POST /tools/execute  OR  stdio (MCP)
                         ▼
┌────────────────────────────────────────────────────────────────────┐
│                      FastAPI MCP Server                            │
│                                                                    │
│   GET /tools           →  ToolRegistry (list all 160 tools)        │
│   POST /tools/execute  →  ToolExecutor (validate + run)            │
│   GET /tools/{name}    →  ToolRegistry (describe one tool)         │
│   GET /metrics         →  Prometheus metrics                       │
└──────────┬─────────────────────────────────────────────────────────┘
           │
    ┌──────┴──────────────────────────────────────────────────────┐
    │               Tool Handlers (20 services)                    │
    │  terraform  github   gitlab   aws   kubernetes  helm         │
    │  azure      gcp      argocd   vault  pagerduty  datadog      │
    │  docker     jenkins  cloudflare  ansible  security  finops   │
    └──────┬──────────────────────────────────────────────────────┘
           │
    ┌──────┴──────────────────────────────────────────────────────┐
    │                     Integrations                             │
    │  subprocess(tf/helm/docker/ansible/trivy/tfsec/infracost)   │
    │  boto3  PyGithub  k8s-client  azure-sdk  google-cloud-sdk   │
    │  httpx(argocd/vault/pagerduty/datadog/gitlab/jenkins/cf)    │
    └─────────────────────────────────────────────────────────────┘

devops-mcp-server/
│
├── server/
│   ├── main.py              # FastAPI app — all HTTP routes
│   ├── mcp_stdio.py         # MCP stdio transport for Claude Desktop
│   ├── registry.py          # Tool registration — build_registry()
│   ├── executor.py          # ToolExecutor — validate + dispatch
│   └── schemas.py           # Pydantic request/response models
│
├── core/
│   ├── config.py            # pydantic-settings — all env vars
│   ├── auth.py              # API key middleware + credential helpers
│   ├── logger.py            # structlog structured logging
│   └── audit.py             # SQLite audit trail (WAL mode)
│
├── tools/
│   ├── terraform/           # plan, apply, destroy, init, validate, output, state_list
│   ├── github/              # create_pr, get_repo, list_issues, trigger_workflow,
│   │                        #   create_release, create_issue, merge_pr, get_workflow_run
│   ├── gitlab/              # list_projects, list_mrs, create_mr, merge_mr,
│   │                        #   list_pipelines, trigger_pipeline, list_issues
│   ├── aws/                 # ec2, s3, lambda, rds, ec2_lifecycle, s3_objects,
│   │                        #   cloudwatch, secrets, networking, iam, rds_crud,
│   │                        #   ecs, cost, ecr, alb, sqs, sns, dynamodb
│   ├── kubernetes/          # deploy, get_pods, get_logs, get_events, scale,
│   │                        #   rollout_restart, rollout_status, get_deployments,
│   │                        #   get_services, get_nodes, delete_pod,
│   │                        #   namespace, configmap, secret, jobs, ingress
│   ├── helm/                # list, install, upgrade, rollback, status
│   ├── azure/               # rg_list, vm_list, vm_start, vm_stop,
│   │                        #   aks_list, acr_list, kv_get, kv_set
│   ├── gcp/                 # instances, buckets, gke, cloudrun, cloudsql,
│   │                        #   cloudbuild, secret_manager
│   ├── argocd/              # list, status, sync, rollback
│   ├── vault/               # read, write, list
│   ├── pagerduty/           # list_incidents, acknowledge, resolve, create
│   ├── datadog/             # monitors, metrics, events, dashboards, incidents, hosts
│   ├── docker/              # list_images, pull, build, push, inspect,
│   │                        #   list_containers, logs, tag
│   ├── jenkins/             # list_jobs, get_job, trigger_build, get_build,
│   │                        #   get_build_log, list_builds
│   ├── cloudflare/          # list_zones, dns_records, purge_cache, waf_rules
│   ├── ansible/             # run_playbook, list_hosts, ping, run_module
│   ├── security/            # trivy_scan_image, trivy_scan_filesystem, tfsec_scan
│   └── finops/              # azure_cost, gcp_billing, infracost_estimate
│
├── integrations/
│   ├── terraform_runner.py  # Terraform subprocess wrapper
│   ├── helm_runner.py       # Helm subprocess wrapper
│   ├── docker_runner.py     # Docker CLI subprocess wrapper
│   ├── ansible_runner.py    # Ansible CLI subprocess wrapper
│   ├── scanner_runner.py    # Trivy + tfsec subprocess wrappers
│   ├── aws_client.py        # boto3 client classes (15 services)
│   ├── github_client.py     # PyGithub wrapper
│   ├── gitlab_client.py     # GitLab REST API v4 (httpx)
│   ├── k8s_client.py        # kubernetes-client/python wrapper
│   ├── azure_client.py      # Azure SDK client classes + Cost Management
│   ├── gcp_client.py        # GCP SDK client classes + Secret Manager + Billing
│   ├── argocd_client.py     # ArgoCD REST API (httpx)
│   ├── vault_client.py      # Vault KV v2 REST API (httpx)
│   ├── pagerduty_client.py  # PagerDuty REST API v2 (httpx)
│   ├── datadog_client.py    # Datadog REST API v1/v2 (httpx)
│   ├── jenkins_client.py    # Jenkins REST API (httpx)
│   └── cloudflare_client.py # Cloudflare REST API v4 (httpx)
│
├── tests/
├── Dockerfile
├── docker-compose.yml
├── pyproject.toml           # PyPI packaging (v4.0.0)
├── requirements.txt
└── .env.example

git clone https://github.com/MaripeddiSupraj/devops-mcp-server.git
cd devops-mcp-server
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

cp .env.example .env
# Fill in credentials for the services you want to use

uvicorn server.main:app --reload

curl http://localhost:8000/tools | python -m json.tool | grep '"count"'

curl -X POST http://localhost:8000/tools/execute \
  -H "Content-Type: application/json" \
  -d '{"tool_name": "aws_list_ec2_instances", "inputs": {}}'

uvicorn server.main:app --host 0.0.0.0 --port 8000

python -m server.mcp_stdio
# or after pip install:
devops-mcp-server

{
  "mcpServers": {
    "devops": {
      "command": "devops-mcp-server",
      "env": {
        "GITHUB_TOKEN": "ghp_...",
        "GITLAB_TOKEN": "glpat-...",
        "AWS_ACCESS_KEY_ID": "AKIA...",
        "AWS_SECRET_ACCESS_KEY": "...",
        "AWS_REGION": "us-east-1",
        "KUBECONFIG": "/home/user/.kube/config",
        "AZURE_SUBSCRIPTION_ID": "...",
        "AZURE_TENANT_ID": "...",
        "AZURE_CLIENT_ID": "...",
        "AZURE_CLIENT_SECRET": "...",
        "GCP_PROJECT_ID": "my-project",
        "GOOGLE_APPLICATION_CREDENTIALS": "/home/user/gcp-key.json",
        "DATADOG_API_KEY": "...",
        "DATADOG_APP_KEY": "...",
        "JENKINS_URL": "https://jenkins.example.com",
        "JENKINS_USER": "admin",
        "JENKINS_TOKEN": "...",
        "CLOUDFLARE_API_TOKEN": "...",
        "ARGOCD_SERVER_URL": "https://argocd.example.com",
        "ARGOCD_AUTH_TOKEN": "...",
        "VAULT_ADDR": "https://vault.example.com",
        "VAULT_TOKEN": "...",
        "PAGERDUTY_API_KEY": "..."
      }
    }
  }
}

docker build -t devops-mcp-server .
docker run --env-file .env -p 8000:8000 devops-mcp-server
# or
docker-compose up

# 1. Get alerted
incidents = execute_tool("pagerduty_list_incidents", {"status": "triggered"})
execute_tool("pagerduty_acknowledge_incident", {"incident_id": "P1234AB"})

# 2. Check Datadog — what's firing?
execute_tool("datadog_list_monitors", {"status": "Alert"})
execute_tool("datadog_query_metrics", {"query": "avg:system.cpu.user{env:prod}", "minutes": 30})

# 3. Pull CloudWatch logs
execute_tool("aws_cloudwatch_query_logs", {
    "log_group": "/aws/ecs/api",
    "query": "fields @message | filter @message like /ERROR/ | limit 20",
    "minutes": 15
})

# 4. Check K8s pods
execute_tool("k8s_get_pods", {"namespace": "production"})

# 5. Rolling restart to clear bad state
execute_tool("k8s_rollout_restart", {"deployment": "api", "namespace": "production"})

# 6. Mark deployment event in Datadog
execute_tool("datadog_create_event", {
    "title": "Rollout restart — api", "text": "Auto-restarted by AI agent",
    "tags": ["env:prod", "service:api"]
})

# 7. Resolve
execute_tool("pagerduty_resolve_incident", {"incident_id": "P1234AB"})

# 1. Trigger Jenkins build
execute_tool("jenkins_trigger_build", {"job_name": "api-build", "params": {"VERSION": "v2.1.0"}})

# 2. Watch build log
execute_tool("jenkins_get_build_log", {"job_name": "api-build", "build_number": 42})

# 3. Scan image for CVEs before shipping
execute_tool("trivy_scan_image", {"image": "myrepo/api:v2.1.0", "severity": "HIGH,CRITICAL"})

# 4. Push to ECR
execute_tool("docker_push", {"image": "123456789.dkr.ecr.us-east-1.amazonaws.com/api:v2.1.0"})

# 5. Scan Terraform before apply
execute_tool("tfsec_scan", {"path": "/infra/prod"})
execute_tool("infracost_estimate", {"path": "/infra/prod"})
execute_tool("terraform_apply", {"working_dir": "/infra/prod"})

# 6. Deploy via Helm
execute_tool("helm_upgrade", {
    "release": "api", "chart": "myrepo/api",
    "namespace": "production", "values": {"image.tag": "v2.1.0"}
})

# 7. Sync ArgoCD
execute_tool("argocd_sync_app", {"app_name": "api-production"})

# 8. Create GitHub release
execute_tool("github_create_release", {
    "owner": "my-org", "repo": "api", "tag": "v2.1.0", "name": "v2.1.0"
})

# 9. Purge Cloudflare cache
execute_tool("cloudflare_purge_cache", {"zone_id": "abc123"})

# AWS
execute_tool("aws_cost_by_service", {"start_date": "2026-04-01", "end_date": "2026-04-22"})

# Azure
execute_tool("azure_cost_by_service", {"start_date": "2026-04-01", "end_date": "2026-04-22"})

# GCP
execute_tool("gcp_billing_monthly_spend", {"dataset": "billing", "table": "gcp_billing_export_v1_XXXX"})

# Estimate next Terraform change
execute_tool("infracost_estimate", {"path": "/infra/aws"})

TOOL_NAME = "my_service_my_action"
TOOL_DESCRIPTION = "One-line description for AI agents."
TOOL_INPUT_SCHEMA = {
    "type": "object",
    "properties": {
        "param": {"type": "string", "description": "..."},
    },
    "required": ["param"],
    "additionalProperties": False,
}

def handler(param: str):
    return {"result": param}

from tools.my_service import my_tool

registry.register(ToolEntry(
    name=my_tool.TOOL_NAME,
    description=my_tool.TOOL_DESCRIPTION,
    input_schema=my_tool.TOOL_INPUT_SCHEMA,
    handler=my_tool.handler,
    tags=["my_service"],
))

pip install pytest pytest-asyncio pytest-mock
pytest tests/ -v
# 152 tests, all pass

curl http://localhost:8000/tools | python -m json.tool | grep '"count"'
# Should be 160

aws sts get-caller-identity

kubectl cluster-info

which docker ansible trivy tfsec infracost helm terraform