Top 7 MCP Servers for AWS Infrastructure Automation

AWS is where most engineering organisations’ infrastructure lives, and AWS is increasingly where AI automation delivers the most immediate ROI. The challenge has always been the complexity of the AWS API surface: EC2, RDS, ECS, EKS, Lambda, S3, IAM, CloudWatch, CloudTrail, Cost Explorer, and hundreds of other services each have their own SDKs, CLI syntax, and console navigation patterns. Even experienced AWS engineers spend significant time on the mechanical work of cross-referencing state across services, constructing CLI queries, and navigating the console.

MCP servers for AWS change this. Instead of running aws ec2 describe-instances with the correct — filters syntax, you ask your AI assistant a question in natural language and the MCP server translates it to the correct API call, executes it, and returns the result in context-aware format. Instead of switching between the EC2 console, CloudWatch, CloudTrail, and Cost Explorer to diagnose a cost spike, you ask a single question and the AI queries all four simultaneously.

This guide covers the seven most valuable MCP servers for AWS infrastructure teams in 2026. We go beyond installation instructions to cover the specific workflows each server enables, the IAM permission design you should implement, and the combination patterns that unlock the most powerful multi-service automation.

SECURITY-FIRST NOTE

Every MCP server described in this guide should run with a dedicated IAM role or user with the minimum permissions required for its use cases. Never share credentials between MCP servers. Rotate credentials quarterly. Enable CloudTrail logging and review AI-generated API calls regularly — you are accountable for what runs under your credentials, regardless of whether a human or an AI made the call.

IAM Foundation: Before You Connect Any MCP Server

The right IAM architecture for AWS MCP servers uses dedicated IAM identities with separate permission scopes for investigative (read) and automation (write) use cases. This is not optional — it is the foundation that makes AI-assisted AWS automation safe to run in production.

Terraform: IAM role for read-only AWS MCP investigation

resource “aws_iam_role” “mcp_readonly” {

name = “mcp-ai-readonly”

assume_role_policy = jsonencode({

Version = “2012–10–17”

Statement = [{

Effect = “Allow”

Principal = { AWS = “arn:aws:iam::ACCOUNT:root” }

Action = “sts:AssumeRole”

Condition = {

StringEquals = {

“aws:PrincipalTag/MCPRole” = “true”

}

}

}]

})

}

resource “aws_iam_role_policy_attachment” “mcp_readonly” {

role = aws_iam_role.mcp_readonly.name

policy_arn = “arn:aws:iam::aws:policy/ReadOnlyAccess”

}

Additional cost explorer permission (not in ReadOnlyAccess):

resource “aws_iam_role_policy” “cost_explorer” {

role = aws_iam_role.mcp_readonly.name

policy = jsonencode({

Version = “2012–10–17”

Statement = [{

Effect = “Allow”

Action = [“ce:GetCostAndUsage”, “ce:GetRecommendations”,

“ce:GetCostForecast”, “ce:ListCostAllocationTags”]

Resource = “*”

}]

})

}

The 7 AWS MCP Servers

1 AWS Labs MCP Server (Official) — The comprehensive, official AWS MCP implementation

Maintained By

AWS Labs (github.com/awslabs/mcp) — official Amazon-maintained

Coverage

EC2, S3, EKS, ECS, Lambda, IAM, CloudFormation, CloudWatch, RDS, ElastiCache and growing

Install

pip install awslabs-mcp-server OR uvx awslabs.mcp

Auth

Standard AWS credential chain: env vars, ~/.aws/credentials, instance role, or assumed role

Killer Feature

Natural language AWS CLI: ask ‘which EC2 instances in us-east-1 have been running more than 90 days and have average CPU below 3%?’ — gets structured, actionable results

2026 Status

Active development; new service coverage added monthly; use this as your primary AWS MCP

The AWS Labs MCP server covers the broadest surface area and has the most active maintenance cadence. For most AWS operations, start here. The natural language to API translation handles the most common query patterns well: resource listing with filters, cross-service tagging searches, resource utilization queries, and configuration inspection.

Configuration for Claude Desktop:

{

“mcpServers”: {

“aws”: {

“command”: “uvx”,

“args”: [“awslabs.mcp-server”],

“env”: {

“AWS_PROFILE”: “mcp-readonly”,

“AWS_REGION”: “us-east-1”,

“AWS_DEFAULT_REGION”: “us-east-1”

}

}

}

}

Sample questions the AI can now answer:

‘List all S3 buckets that have public access enabled’

‘Show me all RDS instances not in a Multi-AZ configuration’

‘Which Lambda functions have not been invoked in the last 30 days?’

‘Find all security groups that allow 0.0.0.0/0 on port 22’

2 CloudWatch MCP Server — Metrics, alarms, logs, and observability queries

Maintained By

Community: github.com/aws-cloudwatch-mcp (most maintained fork)

Coverage

CloudWatch Metrics, Alarms, Log Insights, Contributor Insights, ServiceLens

Install

npm install -g mcp-server-cloudwatch

Key Tools

query_metrics, run_log_insights, list_alarms, get_alarm_history, get_anomaly_detector

Killer Workflow

‘Run a Log Insights query across all Lambda function logs in the last 24 hours and show me all ERROR patterns with their frequency and first occurrence’

IAM Permissions

cloudwatch:GetMetricData, cloudwatch:DescribeAlarms, logs:StartQuery, logs:GetQueryResults, logs:FilterLogEvents

CloudWatch Log Insights via MCP — what you can now ask:

‘Find all 5xx errors in the payments-api log group in the last 2 hours’

‘Show me the top 10 slowest Lambda invocations this week with their trace IDs’

‘Are there any alarms currently in ALARM state across all regions?’

‘Query the VPC flow logs for any traffic denied to the database subnet group’

Log Insights query the AI generates and executes:

fields @timestamp, @message, @logStream

| filter @message like /ERROR/

| stats count(*) as error_count by bin(5m)

| sort error_count desc

| limit 20

3 AWS Cost Explorer MCP Server — Cloud cost analysis, recommendations, and forecasting

Maintained By

Community + AWS SDK wrapper: github.com/aws-cost-mcp

Coverage

Cost and Usage Reports, Savings Plans, Reserved Instance recommendations, Cost Anomaly Detection

Install

pip install aws-cost-mcp-server

Key Tools

get_cost_by_service, get_cost_anomalies, get_savings_plan_recommendations, get_ri_utilization, get_cost_forecast

Killer Workflow

‘Which teams have exceeded their monthly cloud budget, broken down by service, compared to last month?’

IAM Permissions

ce:GetCostAndUsage, ce:GetRecommendations, ce:GetCostForecast, ce:GetAnomalies (all read-only)

Best Use

Monthly FinOps review: ‘Summarise this month’s AWS spend vs. last month, flag any anomalies, and list the top 3 cost reduction opportunities’

4 AWS IAM MCP Server — IAM inspection, policy analysis, and access investigation

Maintained By

Community: github.com/iam-mcp-server

Coverage

IAM Users, Roles, Policies, Access Advisor, Credential Reports, Permission Boundaries

Install

uvx iam-mcp-server

Key Tools

list_roles, get_role_policies, get_access_advisor, generate_credential_report, simulate_policy, find_unused_permissions

Killer Workflow

‘Which IAM roles have admin-equivalent permissions but have not been used in the last 90 days?’

IAM Permissions

iam:Get, iam:List, iam:GenerateCredentialReport, iam:SimulatePrincipalPolicy

Security Value

Continuous IAM hygiene: ‘Show me all users with console access but no MFA enabled, and all roles with wildcard resource ARNs in their attached policies’

5 Amazon EKS MCP Server — EKS cluster management and Kubernetes-AWS integration

Maintained By

AWS Labs: github.com/awslabs/mcp/eks

Coverage

EKS Cluster management, managed node groups, Fargate profiles, add-on management, IRSA

Install

pip install awslabs-eks-mcp-server

Key Tools

list_clusters, get_cluster, update_nodegroup, list_addons, get_addon, list_fargate_profiles

Killer Workflow

‘Show me all EKS clusters not running the latest supported Kubernetes version, with their current version and the upgrade path’

Combination Power

Use with Kubernetes MCP server: EKS MCP for AWS-level cluster management (node groups, IRSA, add-ons); Kubernetes MCP for workload-level operations (pods, deployments, services)

6 AWS CloudTrail MCP Server — API audit logs, compliance investigation, and change tracking

Maintained By

Community: github.com/cloudtrail-mcp

Coverage

CloudTrail Event History, Trails, Insights, Lake Query

Install

npm install -g mcp-server-cloudtrail

Key Tools

search_events, get_event_history, query_cloudtrail_lake, get_insights

Killer Workflow

‘Show me all IAM changes made in the last 24 hours and who made them, ordered by the risk level of the change’

Incident Value

Root cause investigation: ‘What AWS API calls were made in the 30 minutes before the S3 bucket became public yesterday evening?’

IAM Permissions

cloudtrail:LookupEvents, cloudtrail:GetTrailStatus, cloudtrail:DescribeTrails, cloudtrail:StartQuery (Lake)

7 AWS Systems Manager MCP Server — SSM Parameter Store, Run Command, and fleet management

Maintained By

AWS Labs: github.com/awslabs/mcp/ssm

Coverage

Parameter Store, Secrets Manager integration, Run Command, Session Manager, Patch Manager

Install

pip install awslabs-ssm-mcp-server

Key Tools

get_parameter, put_parameter, list_parameters, send_command, get_command_status, list_inventory

Killer Workflow

‘List all SSM parameters in the /production/ path that have not been rotated in more than 90 days’

Automation Power

‘Run a shell command across all EC2 instances tagged Environment=staging to check their disk utilisation’ — executes via Run Command without SSH access

Security Note

Restrict put_parameter and send_command to specific target groups and parameter paths via IAM conditions. These are write operations requiring careful governance.

Multi-Server AWS Workflows

Workflow 1: Complete Security Posture Review

Active servers: AWS Labs + IAM + CloudTrail MCPs

Ask: ‘Run a security posture review of our AWS account’

AI will:

1. [AWS MCP] Find all security groups with unrestricted ingress (0.0.0.0/0)

2. [AWS MCP] Find all S3 buckets with public access

3. [IAM MCP] Find IAM users without MFA enabled

4. [IAM MCP] Find roles with wildcard (*) resource in policies

5. [CloudTrail MCP] Check for any root account API calls in the last 30 days

6. [AWS MCP] Find RDS instances not encrypted at rest

7. Synthesise findings into a prioritised remediation list

Workflow 2: Automated FinOps Weekly Review

Active servers: AWS Labs + Cost Explorer + CloudWatch MCPs

Ask: ‘Generate my weekly cloud cost review’

AI will:

1. [Cost Explorer MCP] Get this week’s spend by service vs. last week

2. [Cost Explorer MCP] Get any cost anomalies detected

3. [AWS MCP] List EC2 instances with < 5% CPU (idle candidates)

4. [CloudWatch MCP] Verify utilisation metrics for flagged instances

5. [Cost Explorer MCP] Get top savings plan recommendations

6. Generate structured weekly report with actions prioritised by savings impact

Workflow 3: Incident Root Cause via CloudTrail

Active servers: CloudTrail + AWS Labs + CloudWatch MCPs

Ask: ‘The production RDS cluster became unreachable at 14:23. What happened?’

AI will:

1. [CloudTrail MCP] Search events around 14:23 affecting the RDS cluster

2. [CloudTrail MCP] Find any IAM changes, VPC changes, or security group modifications

3. [AWS MCP] Describe current vs. known-good security group rules

4. [CloudWatch MCP] Query RDS database connections and error metrics

5. Correlate: ‘A security group rule was modified at 14:21 by arn:aws:iam::…:role/terraform-apply’

6. Provide specific remediation: which rule to add back

IMPLEMENTATION SEQUENCE

Week 1: AWS Labs MCP + read-only IAM role. Use it for every AWS investigation for one week. Week 2: Add Cost Explorer MCP. Run your weekly cost review through AI. Week 3: Add CloudTrail MCP. Use it for your next security review. Week 4: Add CloudWatch MCP. The combination of these four covers 90% of daily AWS operational queries.