The Best Cloud Platform MCP Servers in 2026

Cloud MCP servers expose a provider's APIs and docs to AI agents over the Model Context Protocol, so Claude Code, Cursor and other hosts can list resources, read logs and deploy. Every major cloud ships one in 2026, and most of them call the provider's API with the credentials on the machine.

This guide compares the main cloud MCP servers on the points teams usually weigh when picking one, and includes Encore's MCP servers for apps built on AWS and GCP.

What to compare:

• Coverage: which services, docs or CLI commands the server exposes.
• Read and write actions: whether the agent can only inspect resources or also change them, and whether a read-only mode exists.
• Auth model: whose credentials the server uses, and how access is limited.
• Local or remote: whether it runs on your machine or as a hosted endpoint.
• Maturity: how long it has been available and how actively it is maintained.

Cloud MCP servers at a glance

AWS MCP servers

AWS

AWS's MCP servers, covered by InfoQ when the ECS, EKS and serverless servers shipped, are an open source collection of over sixty servers. The AWS API MCP Server runs any AWS CLI command through a call_aws tool, the Documentation and Knowledge servers answer questions from AWS docs, and service servers cover ECS, EKS, Lambda, CloudWatch, DynamoDB, IAM, pricing and infrastructure as code. They run locally with uvx:

{
  "mcpServers": {
    "aws-api": {
      "command": "uvx",
      "args": ["awslabs.aws-api-mcp-server@latest"],
      "env": {
        "AWS_REGION": "us-east-1",
        "READ_OPERATIONS_ONLY": "true"
      }
    }
  }
}

The servers use the AWS credentials on the machine, so the IAM role defines what the agent can do. READ_OPERATIONS_ONLY blocks write operations on the API server, and AWS describes IAM as the primary control. Because of that breadth, one server can touch every service the role allows.

Google Cloud Run MCP and gcloud MCP

Google Cloud

Google's Cloud Run MCP server deploys local folders or file contents to Cloud Run and lists services and logs, with tools such as deploy-local-folder, list-services and get-service-log. It runs with npx -y @google-cloud/cloud-run-mcp and the gcloud credentials on the machine. The gcloud MCP server exposes gcloud CLI commands more broadly, with configuration to deny specific commands.

Cloud Run MCP is focused enough to be a good default for agents that deploy prototypes. It deploys containers and source, and databases, Pub/Sub and IAM bindings are still set up separately.

Azure MCP Server

Azure

The Azure MCP Server is Microsoft's server for Azure services such as Storage, Cosmos DB, Key Vault, Azure Monitor and resource groups, started with npx -y @azure/mcp@latest server start and authenticated through the Azure CLI or other Azure identity sources. Access follows Azure RBAC on the signed-in identity. It is the default for Azure, and outside this guide's AWS and GCP focus.

Cloudflare MCP servers

Cloudflare runs product-specific MCP servers as hosted endpoints, such as bindings.mcp.cloudflare.com for Workers bindings, a builds server and an observability server, with OAuth against your Cloudflare account. Hosted servers avoid local credentials, and each server is limited to one product area, which keeps tool lists short.

Terraform MCP Server

HashiCorp's Terraform MCP Server gives agents current provider and module documentation from the Terraform Registry, which reduces invented resource arguments, and with an HCP Terraform token it can work with workspaces and runs. It runs as a Docker image. It helps agents write HCL and does not plan or apply on its own machine, so the review and apply workflow stays in place.

Encore MCP servers

Encore is a backend framework and platform where infrastructure is declared in code, and its MCP servers describe one application rather than a whole cloud account. The local MCP server starts with encore mcp start and offers tools such as get_services, query_database, call_endpoint and get_traces, so an agent can call an endpoint it changed and read the resulting trace.

The cloud MCP server exposes traces, deployment status and environment metadata for deployed environments, with OAuth or an API key. It does not change cloud resources directly; those change through code and a deploy to your AWS account or GCP project. The coverage is limited to Encore apps, so it does not replace a provider server for resources outside the app. AI integration covers editor rules for agents.

Credentials and prompt injection

The MCP specification's security best practices cover token handling, session hijacking and confused-deputy problems for servers that proxy other APIs. The bigger practical risk is the combination Simon Willison calls the lethal trifecta: access to private data, exposure to untrusted content such as logs or issues, and a way to send data out. A cloud MCP server with broad credentials often supplies all three, so use read-only modes, separate roles for agents and keep production write access out of agent sessions.

How to choose

• AWS MCP servers for agents that need AWS docs or to inspect AWS resources, with read-only mode and a scoped role.
• Cloud Run MCP for agents deploying prototypes to GCP, and gcloud MCP for broader GCP access.
• Azure MCP Server for Azure.
• Cloudflare MCP servers for Workers and Cloudflare products.
• Terraform MCP Server for agents writing Terraform.
• Encore MCP servers for agents working on an Encore app, locally and in its deployed environments.

Related guides

• Best infrastructure for AI coding agents
• Best local development setup for AI coding agents
• CLAUDE.md and AGENTS.md for AWS deploys
• Best secrets management tools