MCP DevOps Servers: Cloud, K8s, IaC, CI/CD (2026)

DevOps servers connect AI agents to the infrastructure that teams build, deploy, and operate. Independent evaluations of every significant server in this category, covering infrastructure as code, container orchestration, cloud platforms, and deployment pipelines, not directory listings or vendor documentation.

Model Context Protocol (MCP) lets AI clients deploy workers, query clusters, inspect state, trigger pipelines, and manage cloud resources through a standardized JSON-RPC 2.0 interface. Anthropic introduced MCP as an open standard in November 2024. The Linux Foundation’s Agentic AI Foundation (AAIF) now governs the specification (Linux Foundation, “Agentic AI Foundation Launch,” December 9, 2025). 8 servers compete in this category as of mid-2026, and 6 provide official vendor-maintained implementations.

The full MCP server directory covers all 14 server categories. This page focuses on DevOps and infrastructure servers alone.

What Does a DevOps MCP Server Do?

A DevOps MCP server exposes infrastructure management, container orchestration, cloud platform operations, or deployment pipeline control as callable tools. The AI client deploys workers, queries clusters, inspects state, and triggers pipelines through a standardized JSON-RPC interface.

DevOps servers carry higher stakes than most MCP categories. A file system server reads local directories. A communication server sends messages. A DevOps server modifies live infrastructure. The permission model and safety controls matter more here than in any other category.

4 infrastructure domains define this category:

Infrastructure as Code (IaC) servers connect to Terraform, Pulumi, and similar platforms. Tools expose workspace management, run triggers, state inspection, cost estimation, and registry browsing. The AI client plans and applies infrastructure changes through the same workflows teams already use.
Containers and Kubernetes servers connect to Docker, Kubernetes, OpenShift, and orchestration platforms. Tools expose cluster queries, pod management, root cause analysis, and container operations. The AI client investigates failing pods and correlates logs with events.
Cloud platform servers connect to Cloudflare, AWS, Azure, and similar providers. Tools expose service-specific operations across compute, storage, networking, and security. The broadest tool surfaces in the MCP ecosystem live in this domain.
CI/CD servers connect to Vercel, GitHub Actions, and deployment platforms. Tools expose deployment management, environment variable configuration, and pipeline triggers.

Each domain carries different risk profiles. IaC servers modify infrastructure definitions. Container servers modify running workloads. Cloud platform servers modify production services. CI/CD servers trigger deployments. Every domain requires scoped permissions.

How Do You Evaluate DevOps Servers?

Evaluate DevOps servers on 4 criteria: infrastructure domain, production safety controls, vendor maintenance status, and scope control granularity. These dimensions determine whether a server is safe for production infrastructure.

Infrastructure Domain

IaC servers manage definitions. Container servers manage workloads. Cloud servers manage services. CI/CD servers manage deployments. Match the domain to the infrastructure the team operates. Most teams install 2 to 3 DevOps servers covering different domains.

Production Safety Controls

Read-only mode availability is essential. The ability to inspect infrastructure without modifying it separates safe exploration from risky automation. Terraform MCP supports read-only toolsets (registry browsing only). Cloudflare MCP scopes API tokens per service. Kubernetes MCP inherits kubeconfig permissions.

Vendor Maintenance Status

Cloudflare, Docker, HashiCorp (Terraform), AWS, Microsoft (Azure DevOps), and Lens provide official servers. Vercel relies on community implementations. Official servers receive security patches and API compatibility updates. Community servers depend on volunteer maintenance.

Scope Control Granularity

API token scoping determines what the agent can access. Cloudflare supports per-service token scoping (a token for Workers only, a separate token for DNS only). AWS MCP inherits IAM policies. Terraform MCP separates registry browsing from workspace management via toolset flags.

The rubric above applies to every server profiled in the next section.

Which Servers Connect to DevOps Infrastructure Through MCP?

8 DevOps servers expose MCP-compatible tools as of mid-2026. Cloudflare, Docker, Terraform, AWS, Azure DevOps, and Lens provide official vendor-maintained servers. Cloudflare leads with 16+ specialized servers covering its entire platform.

Cloudflare MCP (Official)

Cloudflare provides 16+ specialized MCP servers plus 1 master API server at mcp.cloudflare.com/mcp. The master server uses OAuth authentication and exposes 2 meta-tools: search() to discover API endpoints and execute() to call them. This covers Workers, KV, R2, D1, Pages, DNS, Firewall, Load Balancers, Stream, Images, AI Gateway, Vectorize, and Access.

The Workers Bindings server exposes 25 tools for managing Workers infrastructure: D1 databases, R2 object storage, and KV stores. Scoped API tokens restrict access per service.

Cloudflare occupies a unique dual role in the MCP ecosystem. The platform provides MCP servers for managing Cloudflare resources and also serves as a hosting platform for deploying custom MCP servers on Workers with global edge distribution.

Docker MCP Toolkit (Official)

Docker provides the MCP Toolkit through Docker Desktop. The catalog at hub.docker.com/mcp distributes 200+ MCP servers that run directly in Docker containers. The Toolkit handles OAuth authentication automatically for remote servers.

Docker MCP Toolkit functions as an ecosystem distribution platform. Instead of installing individual servers via npm or pip, teams browse the Docker MCP Catalog and add servers to their Docker Desktop profile with one-click installation. Container isolation provides security boundaries between servers.

Kubernetes MCP (containers/kubernetes-mcp-server)

The containers/kubernetes-mcp-server supports Kubernetes and OpenShift clusters. Available as a native binary, npm package, Python package, and container image. The server provides full cluster visibility through natural language queries.

Root cause analysis correlates logs and events to explain failing pods. The agent identifies the issue and proposes or applies fixes. Limitation: the server requires clusters in the local kubeconfig. Cloud provider clusters (EKS, AKS, GKE) must be configured in kubeconfig before the server can access them.

Lens MCP (Official)

Lens provides its own MCP server that extends Kubernetes management beyond the base containers/kubernetes-mcp-server. The key differentiator: Lens integrates natively with AWS EKS and Azure AKS. The Lens MCP server connects directly to cloud provider clusters without kubeconfig workarounds.

Teams already using Lens as their Kubernetes IDE gain MCP capabilities within their existing workflow. The server inherits Lens cluster permissions and context.

Terraform MCP (HashiCorp Official)

HashiCorp maintains the official Terraform MCP server at hashicorp/terraform-mcp-server. Toolsets separate registry browsing from workspace management: --toolsets=registry for read-only registry access, --toolsets=terraform for workspace operations, --toolsets=all for full access.

The server exposes workspace listing, run triggers, state inspection, cost estimation, and registry provider search. Teams using Terraform Cloud manage all infrastructure operations through the AI agent.

The deprecated AWS Terraform MCP server redirects to HashiCorp’s official implementation. Teams finding outdated guides referencing the AWS version need to migrate to hashicorp/terraform-mcp-server.

AWS MCP (Official, Preview)

AWS provides MCP server support in preview mode. The server covers EC2, S3, IAM, CloudWatch, Lambda, ECS, and EKS. Preview status means the API surface and tool definitions are subject to change.

AWS MCP inherits IAM policies for permission scoping. An agent with a read-only IAM role inspects infrastructure without modification. A deployment IAM role enables write operations. The IAM permission model is the most granular in the DevOps category.

Azure DevOps MCP (Microsoft Official)

Microsoft maintains the Azure DevOps MCP server. Tools expose work item management, pull request operations, build pipeline triggers, and test plan execution. The server targets teams using Azure DevOps for project management alongside Azure cloud infrastructure.

Vercel MCP (Community)

Community-maintained Vercel MCP servers connect to the Vercel API. Tools expose deployment management, DNS configuration, domain management, project settings, and environment variable operations. Vercel MCP targets frontend teams deploying Next.js and other frameworks through the Vercel platform.

Server selection depends on the infrastructure stack. Cloudflare MCP covers edge computing and serverless. Docker MCP Toolkit distributes containerized servers. Kubernetes and Lens cover cluster orchestration. Terraform covers IaC. AWS and Azure cover cloud platforms. Vercel covers frontend deployments.

The sections above define DevOps MCP servers, provide an evaluation rubric, and profile every significant server in the category. The sections below cover production access safety practices and answer the questions that surface during infrastructure server selection.

How Do You Scope Production Access Safely?

Start every DevOps MCP integration with read-only permissions. Validate agent behavior on staging environments before granting write access to production. Scope API tokens to the minimum services required. Production write access is the last step, not the first.

Read-only access lets the agent inspect infrastructure, list resources, query logs, and analyze state without modification risk. Terraform MCP’s --toolsets=registry flag limits the agent to registry browsing. Cloudflare’s scoped tokens restrict access per service. AWS IAM policies enforce read-only boundaries.

Staging validation is the second step. Grant write access to a staging environment. Observe how the agent creates, modifies, and deletes resources. Verify that operations match expectations before promoting the same permissions to production.

Production write access requires the narrowest possible scope. A Terraform token scoped to one workspace. A Cloudflare token scoped to one zone. An AWS IAM role scoped to one service.

Frequently Asked Questions

Which DevOps Server Has the Broadest Tool Surface?

Cloudflare MCP provides the broadest tool surface with 16+ specialized servers and a master API server covering the entire Cloudflare platform. The Workers Bindings server alone exposes 25 tools. By comparison, Terraform MCP exposes registry, workspace, and state tools across 3 configurable toolsets.

Can Docker MCP Toolkit Replace Individual Server Installations?

Docker MCP Toolkit distributes 200+ servers through Docker Desktop’s MCP Catalog. Teams browse and install servers without managing npm, pip, or individual configuration files. Container isolation provides security boundaries. The Toolkit is an alternative installation path, not a replacement for the servers themselves.

What Is the Difference Between Kubernetes MCP and Lens MCP?

Kubernetes MCP (containers/kubernetes-mcp-server) requires clusters in the local kubeconfig. Cloud provider clusters (EKS, AKS) must be manually configured. Lens MCP integrates natively with AWS EKS and Azure AKS, connecting to cloud clusters directly. Teams already using Lens as their Kubernetes IDE benefit from evaluating Lens MCP first.

Is AWS MCP Ready for Production?

AWS MCP is in preview mode as of mid-2026. The API surface and tool definitions are subject to change. Preview status means AWS has not committed to backward compatibility. Teams can use AWS MCP for exploration and development workflows today. Production infrastructure management through AWS MCP carries risk until general availability. The preview designation also means AWS has not yet finalized the IAM permission model specific to MCP access patterns.

Which MCP Servers Connect to DevOps Infrastructure?