Five MCP servers to rule the cloud

Anthropic’s Model Context Protocol (MCP), coined the “USB-C for AI,” has inspired the software industry to think bigger with their AI assistants. Now, armed with access to external data and APIs, as well as to internal platforms and databases, agents are getting arms and legs to conduct impressive automation.

MCP is no longer reserved for trendy AI startups or niche software-as-a-service providers, as the major clouds have begun experimenting with adding MCP servers to their offerings to help customers automate core cloud computing operations. These MCP servers sit alongside and complement existing CLIs and APIs as a protocol for AI consumption.

Using an MCP server connected to the cloud of their choosing, engineers could improve the quality of responses of AI coding agents by providing them with cloud documentation and other highly contextual information relevant to the host cloud. But it goes further than knowledge gathering. MCP in cloud computing could be used to spin up new servers and adjust configurations, or retrieve production metadata in an instant.

The best part is that any MCP-compliant AI client should be able to interact with these servers using nothing more than natural language commands. Popular AI-powered development environments, such as Cursor, Windsurf, and Visual Studio Code, and LLM-driven AI agents, such as Claude, Codex, and GitHub Copilot, all support MCP out of the box.

Below, we’ll examine MCP servers from the major cloud providers. These officially supported remote MCP servers use your existing cloud credentials to enable authenticated API calls from AI clients. They’re free to use in the sense that no additional licensing is required, beyond the standard cloud service and data transfer costs they generate.

AWS MCP servers

Amazon Web Services (AWS) provides a suite of over 60 official MCP servers that span the wide AWS product catalog. AWS MCP servers run the gamut—from servers that provide access to documentation, to those dedicated to infrastructure and deployment, containers, Lambda functions, AI/ML frameworks, data and analytics, messaging, cost analysis, and more.

The general-purpose AWS MCP Server is the best place to start. It’s a remote server hosted by AWS that connects agents with the latest documentation, API references, and standard operating procedures (SOPs) to execute multi-step workflows. These SOPs can configure and provision infrastructure, as well as monitor and analyze cloud costs.

To consider a real-world MCP use case, take troubleshooting an error that has affected multiple AWS services. You might prompt the AWS MCP Server with a command like, “Investigate increased 5xx errors in prod over the last 30 minutes.” Paired with the right context and permissions, the AWS MCP Server will access relevant metrics, logs, and configuration data across services to surface a likely root cause.

A major benefit of AWS’s approach to MCP is that the servers are officially maintained. The catalog is both comprehensive and continuing to evolve, including a gradual migration toward Streamable HTTP, an improved transport protocol. Overall, AWS has clearly invested heavily in MCP as a foundation for agent-forward cloud operations.

Azure MCP Server

Microsoft Azure’s Azure MCP Server allows AI agents to interact with Azure services via natural language commands. Instead of providing separate MCP servers, Azure breaks its MCP server into more than 40 individual MCP tools that span Azure best practices, AI/ML services, analytics, compute, containers, databases, devops, IoT, storage, and other categories.

Using the Azure MCP Server, you can interact conversationally with Azure using prompts such as “Show me all my resource groups” or “List blobs in my storage container named ‘documents,’” according to the documentation. Other queries can list databases, enumerate Azure storage accounts, analyze large datasets in Azure databases, and perform plenty of other actions.

Azure provides an easy-to-follow getting-started guide, with a bit more hand-holding compared to AWS. The documentation clearly walks through installation, tool parameters, and settings to enable or disable agent control over sensitive functions. Each tool has solid documentation with examples of possible natural language prompts.

Google Cloud MCP servers

Google Cloud Platform (GCP) announced its official Google Cloud MCP servers in December 2025. As such, Google Cloud remote MCP servers are still in preview at the time of writing, meaning they are available “as is” with limited support. Nonetheless, Google Cloud currently provides four official remote MCP servers that are operational, spanning dataset operations, virtual machine management, Kubernetes management, and more.

For example, a natural language expression like “Get metadata details and table IDs my dataset ‘newUsers’” issued to the BigQuery MCP server would likely invoke tools like list_table_ids to list table IDs, along with get_dataset_info and get_table_info to retrieve metadata.

Alternatively, you could issue a command like “Kill my running VM in project 0009 in the east zone” to the Compute Engine MCP, which could invoke the stop_instance tool to stop the VM. Other tools support actions like deletion or resetting instances, as well as more benign commands like getting compute metadata and operational traces.

Google Cloud provides MCP servers for Google Kubernetes Engine (GKE) and Google Security Operations. Google also offers the Maps Grounding Lite MCP server, which helps developers build LLM apps on the Google Maps Platform, along with a number of other open-source servers intended for local hosting.

Similar to other cloud offerings, Google Cloud MCP servers provide controls to enable read-only or read-write functions. One unique benefit is Google’s approach to logging for all MCP interactions and access, which could help auditing for cloud administrators. Although GCP currently offers a much sparser array of MCP servers than the other hyperscalers, its MCP tools are promising for automating core cloud computing operations.

Oracle MCP servers

Oracle has a long history of providing private and public cloud options for enterprises. More recently, it has dipped a toe into MCP with a small set of MCP servers that wrap popular Oracle platforms. These servers can manage Oracle Cloud Infrastructure (OCI) and operate on Oracle databases and MySQL resources.

For example, Oracle SQLcl is the command-line interface (CLI) for Oracle Database, and its MCP server enables agents to execute queries and process results. On the Oracle blog, engineers suggest the prompt, “Connect to my fun side project and tell me about what kind of data I have there,” which invokes a list-connections tool that returns all saved Oracle connections in storage.

Other use cases for Oracle’s MCP servers include describing database schemas in plain language and generating them, analyzing MySQL usage patterns in real time, or pointing a project to existing database tables to populate data within an application.

Some of Oracle’s MCP work remains in a proof-of-concept phase, but it signals an interesting direction for combining well-established database platforms with emerging AI-driven prototyping and development workflows.

IBM Cloud MCP servers

The IBM Cloud MCP servers are experimental at the time of writing, yet they are designed to be a comprehensive knowledge-gathering layer between AI assistants and the IBM Cloud platform. They can be used to retrieve information about services within a user’s IBM Cloud computing environments.

Unlike most MCP servers on this list, which run in the cloud, IBM’s Core MCP Server is intended to be installed locally and then pointed at the IBM Cloud CLI. It’s essentially a layer over the IBM Cloud CLI. The Core MCP Server can also be containerized for fit-for-purpose needs. However, there are a few potential hindrances: the server is stateful, it is not designed for multi-account use, and it does not support OAuth.

Still, the Core MCP Server could be a user-friendly way to query IBM Cloud to discover cloud resources, retrieve extensive metadata, filter results based on strings, list service names, and more. The documentation suggests simple prompts such as “Are there any VPCs in this account?”, “What zones are available in us-east?”, and “What resource groups are in my account?”

Beyond the Core MCP Server, IBM Cloud also provides MCP servers for Cloud Internet Services (DNS, GLB, WAF, DDoS, and CDN), logs, streams, Kubernetes and OpenShift, code monitoring, object storage, serverless services, VPC, and other IBM Cloud services.

Documentation is thorough, with solid examples, and using MCP should feel natural to those already familiar with IBM’s dense CLI and API commands. However, most actions available through IBM Cloud MCP servers are read-only. For the time being, IBM’s MCP servers serve mainly as an experimental, information-gathering interface.

The cloud is your oyster

MCP has gained increasing enterprise traction in recent months, aligning with the emergence of hyperscaler support for the protocol. Used in a cloud operational context, MCP could eliminate tedious tasks like configuring fields in human-facing GUIs or manually searching through API references and product documentation.

Using the MCP servers outlined above presents an exciting prospect: a new, streamlined, AI-driven control layer for operating the hyperscale clouds. At least, that’s the goal. The reality is that it’s still early days, and many of these servers remain in an experimental or preview phase. The security models also vary significantly from server to server, and not all support mutating operations, with many defaulting to read-only modes.

Taking advantage of AI agents, MCP, and natural language to automate cloud operations will require plenty of experimentation and hands-on testing, not to mention creativity. From database lookups and resource management to provisioning, scaling, root-cause analysis, and cost optimization, it’s ultimately up to operators to decide how MCP fits into their workflows. In essence, with MCP, the cloud is your oyster. What will you do with it?