The server provides read-only, evidence-backed Cloud DevOps analysis and risk assessment for infrastructure, identity, Kubernetes, CI/CD, SRE, and supply chain.
Assess Terraform changes and security posture from plan JSON.
Correlate Terraform, IAM, Kubernetes, and GitHub Actions evidence into a deployment-risk bundle.
Build incident runbooks and calculate SLO error budgets.
Review CI/CD pipelines and GitHub Actions workflows.
Review AWS IAM and AWS/Azure/GCP identity policies.
Review Kubernetes deployment readiness and security hardening.
Review software supply chain using SBOMs, CI pinning, image immutability, signing, and provenance.
Optional: enable allowlisted live multi-cloud reads, observability, tracing/SLO, and OpsChugex private intelligence gateways (root-cause, governance, security posture, FinOps, change blast radius).
Provides a tool to assess Terraform or IaC deployment risk based on changed resource classes and release controls.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Cloud DevOps MCP Server assess deployment risk for terraform changes with IAM and public ingress"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Cloud DevOps MCP Server is a Model Context Protocol v2 server by Alex C. Godwin. It provides evidence-backed Cloud DevOps analysis across infrastructure, identity, Kubernetes, CI/CD, SRE and software supply-chain controls.
The v0.14 line adds an opt-in OpsChugex change-intelligence and blast-radius gateway. The public MCP forwards bounded planned-change, topology and readiness evidence to a host-configured private OpsChugex service and returns change risk, impacted topology nodes, representative blast paths, blockers, warnings and evidence gaps. Proprietary dependency propagation, risk weighting and blocker rules are not included in this public MIT repository.
The v0.13 line adds an opt-in OpsChugex advanced FinOps gateway. The public MCP forwards bounded cloud-cost, utilization and Kubernetes allocation evidence to a host-configured private OpsChugex service and returns optimization opportunities, cost anomalies, low/high savings ranges, confidence, cost correlations and evidence gaps. Proprietary savings factors, anomaly thresholds, prioritization and deduplication are not included in this public MIT repository.
The v0.12 line adds an opt-in OpsChugex cloud security posture gateway. The public MCP forwards bounded cloud asset, identity, secret and network-reachability evidence to a host-configured private OpsChugex service and returns security score, risk level, ordered findings, attack paths, evidence gaps and recommendations. Proprietary security rules, severity thresholds, scoring and attack-path correlation are not included in this public MIT repository.
The v0.11 line adds an opt-in OpsChugex policy and governance gateway. The public MCP sends bounded resource evidence and approved exception metadata to a host-configured private OpsChugex service and returns pass, review or block decisions with control findings and evidence gaps. Proprietary profiles, rules, weights and exception-processing logic are not included in this public MIT repository.
The v0.10 line adds an opt-in OpsChugex root-cause intelligence gateway. The public MCP sends bounded incident evidence to a host-configured private OpsChugex service and returns evidence-ranked probable causes, contradictions, limitations and next checks. Proprietary ranking and correlation rules are not included in this public MIT repository.
The v0.9 line adds an opt-in distributed-tracing and SLO-intelligence plane for Grafana Tempo and Jaeger v3 trace reads, service dependency mapping, tracing coverage assessment, multi-window SLO burn-rate analysis and trace/SLO incident correlation. The v0.8 production-observability and operations-intelligence plane remains available for Prometheus, Grafana, CloudWatch Logs Insights, Kubernetes health, GitHub Actions diagnosis, cloud health, FinOps and drift. Live AWS, Azure and GCP access remains bounded and read-only. Cloud mutation remains intentionally unavailable.
Related MCP server: salesforce-deployment-guard
AI assistants are more useful in engineering work when they can call focused tools with clear inputs and consistent outputs. This server provides a Cloud DevOps tool layer for:
Cross-domain release-risk correlation across infrastructure, identity, runtime and delivery.
Infrastructure-as-code deployment risk analysis.
Production incident runbook generation.
CI/CD delivery readiness review.
SLO error budget calculations.
AWS IAM least-privilege review.
AWS, Azure and GCP identity policy packs.
Terraform destructive-change, public exposure and encryption security analysis.
Kubernetes workload production readiness and security-policy analysis.
GitHub Actions workflow security and deployment review.
CycloneDX/SPDX SBOM quality and software supply-chain correlation.
Optional authenticated Streamable HTTP serving for self-hosted remote access.
Optional allowlisted live AWS/Azure/GCP inventory, observability, FinOps and drift signals.
Optional Grafana Tempo and Jaeger v3 trace reads, service dependency maps and SLO burn-rate intelligence.
Optional OpsChugex private root-cause intelligence across metrics, logs, traces, Kubernetes, cloud, Terraform and CI/CD evidence.
Optional OpsChugex private policy and governance intelligence for development, staging, production and regulated profiles.
Optional OpsChugex private cloud security posture intelligence for exposure, identity, secrets and attack-path analysis.
Optional OpsChugex private advanced FinOps intelligence for rightsizing, anomaly detection, Kubernetes/cloud cost correlation and savings-range analysis.
Optional OpsChugex private change intelligence for pre-change dependency propagation, blast-radius mapping, blockers and rollback/readiness analysis.
Tool | Purpose |
| Correlates Terraform, IAM, Kubernetes and GitHub Actions evidence into one deployment-risk assessment with cross-domain change paths. |
| Scores Terraform/IaC risk and can derive evidence from raw Terraform plan JSON. |
| Produces a practical incident response runbook for a service, symptom, environment and severity. |
| Reviews CI/CD maturity while separating failed controls from unknown evidence. |
| Calculates downtime and request-failure budgets with consistency validation. |
| Parses IAM policy JSON and detects wildcard scope and privilege-escalation paths. |
| Parses Kubernetes YAML for probes, resources, disruption protection, image and exposure risks. |
| Parses workflow YAML for triggers, immutable action pins, permissions, caching and concurrency. |
| Applies AWS IAM, Azure RBAC or GCP IAM policy packs to raw policy JSON. |
| Reviews Terraform plan JSON for destructive changes, public exposure, encryption, deletion protection and wildcard IAM. |
| Reviews privileged mode, host access, service accounts, capabilities, seccomp, root filesystems and NetworkPolicy. |
| Correlates CycloneDX/SPDX SBOM quality with CI action pinning, image immutability, signatures and provenance. |
When explicitly enabled, six additional tools provide allowlisted AWS/Azure/GCP identity verification, bounded inventory, managed Kubernetes discovery, observability configuration summaries, FinOps waste signals and drift reporting. No cloud mutation commands are exposed. AWS general inventory is sourced from the Resource Groups Tagging API, so untagged AWS resources may not appear in that inventory or AWS drift comparison.
When explicitly enabled, twelve additional tools provide bounded Prometheus queries, Grafana alert summaries, CloudWatch Logs Insights queries, Kubernetes pod-health summaries, GitHub Actions failure diagnosis, cross-signal incident correlation, cloud-health assessment, deployment/incident correlation, observability coverage assessment, FinOps correlation, cross-runtime drift analysis and operations briefs. Endpoints, cloud scopes, log groups, cluster contexts, namespaces and repositories are allowlisted. The layer is read-only and exposes no alert mutation, deployment mutation or arbitrary shell execution.
When explicitly enabled, six additional v0.9 tools provide bounded Tempo/Jaeger trace search and retrieval, service dependency mapping, tracing coverage assessment, multi-window SLO burn-rate analysis and trace/SLO incident correlation. Remote tracing endpoints must be allowlisted and use HTTPS unless loopback. Backend credentials stay in host environment variables. Trace search windows and result sizes are bounded, and the plane exposes no trace ingestion, sampling mutation or telemetry deletion.
When explicitly enabled, v0.10 exposes diagnose_root_cause. The tool accepts bounded evidence from metrics, logs, traces, Kubernetes, cloud, Terraform and CI/CD, then calls a host-configured private OpsChugex service. The public MCP contains no proprietary ranking rules, accepts no service URL or credential as tool input, and performs no remediation. Evidence scores represent evidence strength rather than statistical probability.
When explicitly enabled, v0.11 exposes assess_governance_policy. The tool accepts bounded factual resource evidence plus optional owner-attributed, time-bounded exceptions and forwards them to the private OpsChugex policy engine. It returns pass, review, or block with control findings and evidence gaps.
The public MCP contains no proprietary governance profiles, policy rules, scoring weights, exception evaluation logic or enforcement capability. The service URL and token remain host-side and cannot be supplied as MCP arguments.
When explicitly enabled, v0.12 exposes assess_cloud_security_posture. The tool accepts bounded asset, identity, secret and network-reachability evidence and forwards it to the private OpsChugex security engine. It returns a security score, risk level, ordered findings, correlated attack paths, evidence gaps and recommended next actions.
The public MCP contains no proprietary security detection thresholds, severity rules, scoring logic or attack-path algorithm. It cannot rotate credentials, change IAM, modify network controls, alter encryption settings or remediate infrastructure.
When explicitly enabled, v0.13 exposes analyze_advanced_finops. The tool accepts bounded cloud-cost, utilization and Kubernetes allocation evidence and forwards it to the private OpsChugex FinOps engine. It returns optimization opportunities, anomalies, savings ranges, confidence, cost correlations and evidence gaps.
The public MCP contains no proprietary savings factors, anomaly thresholds, prioritization rules, confidence algorithm or portfolio deduplication logic. It does not resize resources, terminate workloads, purchase commitments, change Kubernetes requests or perform billing actions.
When explicitly enabled, v0.14 exposes analyze_change_blast_radius. The tool accepts bounded planned-change items, topology nodes/edges and readiness evidence, then forwards them to the private OpsChugex change-intelligence engine. It returns risk level/score, impacted nodes, representative blast paths, blockers, warnings, evidence gaps and recommendations.
The public MCP contains no proprietary graph-propagation algorithm, change-risk weights, production-blocker rules or approval logic. The tool is read-only and cannot apply Terraform, mutate Kubernetes, merge pull requests, deploy workloads, approve changes or execute rollback.
When explicitly enabled, six additional tools provide Terraform format/validation/plan summaries and Kubernetes read-only runtime inspection. These operations use repository, context, namespace and resource allowlists. Full Terraform plan JSON, Kubernetes Secrets, arbitrary shell execution, Terraform apply and Kubernetes mutation are deliberately excluded.
flowchart TD
LocalClient["Local MCP client"] --> Stdio["stdio"]
RemoteClient["Remote MCP client"] --> HTTPS["HTTPS reverse proxy / gateway"]
HTTPS --> AuthHTTP["Bearer-authenticated Streamable HTTP"]
Stdio --> Server["Cloud DevOps MCP server"]
AuthHTTP --> Server
Server --> DomainTools["Domain + policy-pack analyzers"]
DomainTools --> Correlator["Cross-domain and supply-chain correlation"]
DomainTools --> Output["Structured guidance"]
Correlator --> OutputRun the published MCP server directly from npm:
npx -y cloud-devops-mcp-server@0.14.0On Windows PowerShell systems where script execution policy blocks npx.ps1, use:
npx.cmd -y cloud-devops-mcp-server@0.14.0Install the CLI globally if you prefer a persistent local command:
npm install -g cloud-devops-mcp-server@0.14.0
cloud-devops-mcp-serverThe package is published on npm as cloud-devops-mcp-server and registered in the official MCP Registry as io.github.alexcgodwin/cloud-devops-mcp-server.
Cloud DevOps MCP Server supports local stdio clients and MCP clients capable of connecting to Streamable HTTP endpoints. Common local clients include:
Cursor
Claude Desktop
VS Code with MCP support
Claude Code
Other clients that follow the Model Context Protocol stdio transport
Use stdio for normal local operation. For self-hosted remote access, start the optional authenticated Streamable HTTP endpoint and place non-local deployments behind an HTTPS reverse proxy or gateway.
For MCP clients that support local stdio servers, the recommended public configuration is:
{
"mcpServers": {
"cloud-devops": {
"command": "npx",
"args": ["-y", "cloud-devops-mcp-server@0.14.0"]
}
}
}Windows clients can use npx.cmd if npx resolves through a blocked PowerShell wrapper:
{
"mcpServers": {
"cloud-devops": {
"command": "npx.cmd",
"args": ["-y", "cloud-devops-mcp-server@0.14.0"]
}
}
}See docs/configuration.md for npm, global-install, source-development and authenticated Streamable HTTP configuration options.
Local loopback example:
$env:CLOUD_DEVOPS_MCP_BEARER_TOKEN="<random secret at least 32 characters>"
npm run start:httpThe MCP endpoint is http://127.0.0.1:3000/mcp and requires Authorization: Bearer <token>. A non-local bind additionally requires CLOUD_DEVOPS_MCP_ALLOWED_HOSTS and an HTTPS CLOUD_DEVOPS_MCP_PUBLIC_BASE_URL so remote traffic is expected to terminate TLS at a reverse proxy or gateway.
The v0.14.0 release candidate passes 116 automated tests, with 85.80% statement, 72.18% branch, 85.89% function and 89.11% line coverage. The production dependency audit reports zero vulnerabilities. Public clean-install and MCP Registry acceptance are recorded after publication.
See docs/public-acceptance.md for the verification record.
{
"changedResources": ["network", "iam", "kubernetes"],
"includesIamChanges": true,
"includesPublicIngress": true,
"modifiesStatefulResources": false,
"hasRollbackPlan": true,
"hasPeerReview": true,
"hasTerraformPlan": true
}Example output shape:
{
"riskScore": 78,
"riskLevel": "critical",
"changedResources": ["network", "iam", "kubernetes"],
"recommendedReleasePath": "Change-advisory review, maintenance window and staged execution are recommended."
}See docs/demo.md for practical sample inputs and outputs across the toolset.
Build and run the server in a container:
docker build -t cloud-devops-mcp-server .
docker run --rm -i cloud-devops-mcp-servernpm run dev
npm run build
npm test
npm run checkThe core decision logic lives in src/logic.ts and the MCP tool registration lives in src/index.ts.
More project notes are available in DEVELOPMENT.md, RELEASE.md and docs/architecture.md.
Stdio remains the default and requires no secrets.
Optional Streamable HTTP requires a bearer token of at least 32 characters.
Non-local HTTP binds require an explicit Host allowlist and an HTTPS public base URL for reverse-proxy/gateway termination.
Host and Origin validation are enabled through the official MCP Fastify adapter.
The default analysis tools do not require cloud credentials or call cloud APIs.
Optional Terraform/Kubernetes operations may use locally configured provider or cluster credentials after explicit enablement and allowlisting.
Optional live cloud reads use existing AWS CLI, Azure CLI or gcloud authentication and require explicit account/subscription/project allowlists.
Optional production-observability reads require explicit endpoint/resource allowlists and host-managed credentials; returned logs and diagnostics are bounded and redacted.
Optional v0.10 root-cause intelligence uses a host-configured HTTPS endpoint and host-side token; neither is accepted as a tool argument, and the proprietary ranking engine remains outside the public repository.
Optional v0.11 governance intelligence uses a separately gated host-configured HTTPS endpoint and the same host-side OpsChugex token; proprietary policy rules and enforcement remain outside the public repository.
Optional v0.12 cloud security posture intelligence uses its own fail-closed gate and host-configured HTTPS endpoint; proprietary security scoring and attack-path correlation remain in the private OpsChugex core.
Optional v0.13 advanced FinOps intelligence uses its own fail-closed gate and host-configured HTTPS endpoint; proprietary savings, anomaly, prioritization and deduplication logic remain in the private OpsChugex core.
Optional v0.14 change intelligence uses its own fail-closed gate and host-configured HTTPS endpoint; proprietary dependency propagation, risk weighting and blocker logic remain in the private OpsChugex core.
No cloud mutation tool is exposed.
Analysis remains read-only by default. Controlled execution appears only when explicitly enabled and allowlisted.
No generic shell tool or force-push capability is exposed.
Direct commit/push on protected branches is blocked, and high-impact GitHub actions require explicit confirmation.
Analysis outputs are advisory. Optional operational tools remain bounded by explicit allowlists and fixed command/API surfaces.
Current: v0.14.0 Change Intelligence & Blast-Radius Analysis - public gateway to private OpsChugex pre-change dependency propagation, impacted-service mapping, blockers and readiness analysis.
v0.15.0 Controlled Remediation Gateway - approval-gated safe fixes for selected cloud, Kubernetes and Terraform operational problems.
v0.16.0 Multi-Account / Multi-Organization Operations - AWS Organizations, Azure tenants/subscriptions and GCP organizations/projects topology.
v0.17.0 Incident Command & Automated Runbooks - incident timelines, evidence bundles, remediation plans, rollback recommendations and post-incident reports.
v0.18.0 Platform Engineering Intelligence - service catalog, ownership, golden paths, environment health and developer-platform checks.
v0.19.0 Enterprise Authentication & Authorization - OAuth/OIDC, RBAC, per-user scopes, stronger hosted-MCP access controls and audit trails.
v1.0.0 Production Stable Release - stable tool contracts, compatibility guarantees, hardened security model, comprehensive documentation and enterprise-ready release standards.
Built by Alex C. Godwin, Cloud DevOps Engineer.
Correlate evidence from at least two domains (Terraform, IAM, Kubernetes, GitHub Actions) into one deployment-risk assessment and identify cross-domain change paths. Use domain-specific review tools when only one evidence domain is available. It analyzes caller-supplied artifacts only and does not query providers, clusters, GitHub, or deploy changes.
Inputs: changeName (required), environment (required), terraform, iamPolicies, +2 more. Returns: changeName, changePaths, environment, releaseGate, +12 more
Evaluate Terraform/IaC release risk using plan evidence plus change-governance facts such as rollback, peer review and stateful impact. Use this for overall change/release decisions; for security-only inspection of a raw Terraform plan, use review_terraform_security. It analyzes supplied evidence only and never applies a plan, changes infrastructure, or writes Terraform state.
Inputs: hasPeerReview, hasRollbackPlan, changedResources, hasTerraformPlan, +4 more. Returns: evidence, checklist, riskLevel, riskScore, +4 more
Generate a practical incident-response runbook from a known service symptom, severity, environment and optional signals. Use this to structure response actions and evidence collection; do not use it to fetch or diagnose from live telemetry. It does not execute remediation or make changes to the service.
Inputs: service (required), symptom (required), severity (required), environment (required), +1 more. Returns: title, context, mitigation, rcaEvidence, +3 more
Calculate remaining SLO downtime budget and, when request counts are supplied, remaining failed-request budget for a fixed period. Use this as a deterministic budget calculator; do not use it to fetch monitoring data or forecast reliability. It performs no external calls and changes no service state.
Inputs: periodDays (required), sloTargetPercent (required), observedDowntimeMinutes (required), requestVolume, +1 more. Returns: periodDays, budgetStatus, failedRequests, sloTargetPercent, +5 more
Evaluate generic CI/CD production-readiness controls from structured pipeline facts, separating missing evidence from failed controls. Use this for platform-agnostic delivery process review; for raw GitHub Actions YAML, use review_github_actions_workflow. It is analysis-only and does not trigger builds, deployments, approvals, or pipeline changes.
Inputs: environments (required), pipelineName (required), deploymentStrategy (required), hasRollback, +4 more. Returns: findings, strengths, pipelineName, uncertainties, +5 more
Apply deterministic provider-specific identity policy packs to raw AWS IAM, Azure RBAC or GCP IAM policy documents. Use this for cross-cloud identity security analysis; use review_iam_policy when assessing AWS IAM from mixed structured facts or policy JSON. It analyzes supplied policy JSON only and does not call cloud APIs or change permissions.
Inputs: provider (required), policyJson (required), policyName (required), environment. Returns: facts, findings, provider, riskLevel, +10 more
Evaluate GitHub Actions workflow security and deployment readiness from structured facts or raw workflow YAML, including triggers, action pinning, token permissions, caching, environment protection and concurrency. Use review_cicd_pipeline for generic non-GitHub delivery-process review. It examines supplied evidence only and does not call GitHub or dispatch workflows.
Inputs: workflowName (required), triggers, workflowYaml, hasSecretScanning, +6 more. Returns: evidence, findings, triggers, strengths, +6 more
Evaluate AWS IAM policy risk from structured facts or raw policy JSON, including wildcard scope, privilege-escalation actions and conditions. Use this for AWS IAM operational risk; for provider-specific AWS/Azure/GCP policy-pack checks, use review_cloud_identity_policy. It analyzes supplied policy data only and does not call AWS or modify IAM.
Inputs: policyName (required), actions, resources, policyJson, +5 more. Returns: actions, evidence, findings, resources, +7 more
Evaluate Kubernetes workload production readiness and reliability from structured facts or multi-document YAML, including probes, resources, replicas, disruption protection and exposure. Use this for deployability/readiness; for workload security hardening, use review_kubernetes_security. It examines supplied evidence only and does not connect to a Kubernetes cluster.
Inputs: replicas, namespace, runsAsRoot, manifestYaml, +8 more. Returns: evidence, findings, namespace, strengths, +6 more
Apply Kubernetes workload security rules to manifest YAML for privileged mode, host access, Linux capabilities, service accounts, seccomp, filesystem settings and network policy. Use this for security hardening; use review_kubernetes_deployment for reliability and production-readiness checks. It analyzes supplied YAML only and does not connect to a cluster.
Inputs: manifestYaml (required), environment. Returns: findings, riskLevel, riskScore, policyPack, +10 more
Correlate CycloneDX/SPDX SBOM quality with CI action pinning, Kubernetes image immutability, artifact signing and build provenance to assess software-supply-chain risk. Use this when an SBOM is available and supply-chain evidence needs to be evaluated together. It analyzes supplied artifacts only and does not call registries, CI systems, or clusters.
Inputs: sbomJson (required), environment, workflowYaml, hasProvenance, +2 more. Returns: findings, riskLevel, riskScore, policyPack, +16 more
Inspect raw Terraform plan JSON for security-relevant changes such as destructive actions, public exposure, encryption gaps, deletion protection and IAM wildcard risk. Use this for Terraform security posture; use assess_terraform_change for broader release/change risk and governance. It parses the supplied plan only and does not execute Terraform or write state.
Inputs: terraformPlanJson (required), environment. Returns: findings, riskLevel, riskScore, policyPack, +10 more
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 12 tools
Descriptions explicitly cross-reference paired tools (e.g., review_iam_policy vs review_cloud_identity_policy, review_terraform_security vs assess_terraform_change), helping agents distinguish them. However, the set includes several adjacent 'review_' tools (Kubernetes deployment vs security, GitHub Actions vs generic CI/CD) that could still cause hesitation without careful reading.
All tool names follow a consistent snake_case verb_noun pattern using assess_, build_, estimate_, and review_ prefixes. No naming inconsistencies or mixed conventions are present.
With 12 tools, the set sits comfortably within the ideal 3–15 range. Each tool targets a distinct domain or analysis type, and no tool appears redundant.
The surface covers major DevOps analysis domains: Terraform change and security, cloud change bundles, CI/CD and GitHub Actions, IAM and cloud identity, Kubernetes deployment and security, SLO budgets, incident runbooks, and supply chain. Some gaps exist (e.g., cost optimization, compliance, or actual remediation), but the server explicitly scopes itself as analysis-only, so coverage is strong for that purpose.
Read-only AI coding tools for change verification, release readiness, capacity, and guidance.
30+ deterministic dev and delivery tools: linting, metrics, diagrams, model advice, AI-text checks.
Read-only, deterministic AI triage and readiness tools implementing Sophon's published rubrics.
Query your team's drift, vulnerability, and upgrade data from any AI assistant. OAuth 2.1, 51 tools.
