AWS MCP Servers — 54 Active Servers, Nine Deprecated, and Google Just Matched the Count

Fifty-four active MCP servers. Nine deprecated. One monorepo. Two CVEs. And a competitive landscape that just shifted under AWS’s feet.

The AWS MCP servers from awslabs represent a massive investment in Model Context Protocol infrastructure. With 8,900 stars, 1,500 forks, 1,495 commits, and servers spanning everything from S3 storage to genomics workflows, this is AWS betting that natural language will become a primary interface for cloud operations.

Updated April 29, 2026: Stars 8,800→8,900, commits 1,473→1,495, issues 139→147. Major deprecation wave — Cloud Control API, MSK, Code Documentation, CloudWatch AppSignals, Git Research, Bedrock Data Automation all deprecated (9 total). IAM privilege escalation and EKS path traversal security fixes. OpenAPI migrated to FastMCP 3.x. COMPETITIVE LANDSCAPE TRANSFORMED: Google Cloud now has 50 managed remote MCP servers (matching AWS count), Azure ships 230+ tools across 45 services built into VS 2022.

But AWS is no longer alone at the top. Google matched the server count, Azure matched the service breadth, and both did it with zero-install managed approaches. Part of our Cloud & Infrastructure MCP category.

What It Is

Unlike every other review on this site, this isn’t a single server — it’s a suite. The awslabs/mcp monorepo contains 54 active MCP servers organized into categories (down from 68 at initial review — nine servers have been deprecated in a consolidation wave, and 12 deprecated directories were cleaned up in the April 21 release):

Documentation & Knowledge (3 servers):

Infrastructure & Deployment (8 servers):

AI & Machine Learning (9 servers):

Data & Analytics (17 servers):

Developer Tools (2 active servers):

Integration & Messaging (4 servers):

Cost & Operations (8 servers):

Healthcare & Life Sciences (3 servers):

Deprecated servers (9 total, directories cleaned up April 21): Core MCP Server (orchestration layer, March 2026), CDK MCP Server (replaced by IaC), Cloud Control API MCP (replaced by IaC), Terraform MCP (use HashiCorp’s official), MSK MCP (April 2026), Code Documentation MCP, Git Research MCP, CloudWatch AppSignals (folded into CloudWatch), Nova Canvas MCP, and Bedrock Data Automation MCP.

Setup

Note: The Core MCP Server (v1.0.27, deprecated March 2026) is no longer the recommended entry point. AWS now recommends configuring individual servers directly in your MCP client, which supports multi-server setups natively. The Core MCP Server still works but will not receive updates.

The legacy Core MCP Server configuration used role-based environment variables:

{
  "mcpServers": {
    "awslabs-core-mcp-server": {
      "command": "uvx",
      "args": ["awslabs.core-mcp-server@latest"],
      "env": {
        "AWS_PROFILE": "your-profile",
        "AWS_REGION": "us-east-1",
        "SOLUTIONS_ARCHITECT": "enabled",
        "DEV_TOOLS": "enabled"
      }
    }
  }
}

There are 16 predefined roles: aws-foundation, dev-tools, ci-cd-devops, container-orchestration, serverless-architecture, analytics-warehouse, data-platform-eng, frontend-dev, solutions-architect, finops, monitoring-observability, caching-performance, security-identity, sql-db-specialist, nosql-db-specialist, messaging-events, and healthcare-lifesci. Each bundles 2-5 related servers.

Individual servers can also be run standalone:

{
  "mcpServers": {
    "aws-documentation": {
      "command": "uvx",
      "args": ["awslabs.aws-documentation-mcp-server@latest"],
      "env": {
        "FASTMCP_LOG_LEVEL": "ERROR"
      }
    }
  }
}

The AWS Knowledge MCP Server is the easiest to try — it’s a fully managed remote server at https://knowledge-mcp.global.api.aws with Streamable HTTP transport, no authentication required, and no local setup. Just point your MCP client at the URL and start querying AWS documentation.

Setup difficulty: Moderate to Hard. The Knowledge server is one-click. The Documentation server needs Python 3.10+ and uv. The API server and infrastructure servers require AWS credentials, proper IAM permissions, and understanding of which of the 54 active servers you actually need. With the Core MCP Server deprecated, there’s no bundling mechanism — you configure each server individually.

What Works Well

Breadth no longer unmatched, but still deep. AWS has 54 active MCP servers — Google Cloud now matches with 50 managed remote servers, and Azure ships 230+ tools across 45 services. But AWS’s servers go deeper into service-specific operations than either competitor. DynamoDB, EKS, SageMaker, Step Functions, CloudWatch, Cost Explorer — nearly every major AWS service has dedicated MCP tooling with service-specific features that generic cloud APIs don’t expose.

The Knowledge server is genuinely useful. A managed remote endpoint at knowledge-mcp.global.api.aws that indexes AWS documentation, blog posts, What’s New announcements, API references, Getting Started guides, Well-Architected guidance, and CDK/CloudFormation materials — all without authentication. Five tools: search_documentation, read_documentation, recommend, list_regions, and get_regional_availability. The regional availability check alone saves hours of digging through service-specific pages.

The API server’s security model. call_aws validates all CLI commands before execution. READ_OPERATIONS_ONLY restricts to read-only. REQUIRE_MUTATION_CONSENT requires explicit approval before mutating actions. Denylisted services (deploy, emr) are blocked due to subprocess security risks. suggest_aws_commands provides command suggestions based on natural language, including commands released after the model’s knowledge cutoff. This is thoughtful infrastructure safety.

Active development. 1,495 commits, multiple releases per week via automated CI/CD. The April 21 release migrated OpenAPI to FastMCP 3.x, added billing view tools to Cost Management, integrated Runtime tools into AgentCore, and cleaned up 12 deprecated server directories. The AWS API MCP Server alone has 5M+ total PyPI downloads. This is not a side project — it’s a funded initiative with continuous delivery.

CloudTrail audit logging. Every API call made through the managed AWS MCP server is logged to CloudTrail. IAM-based permissions with zero credential exposure. This is how enterprise cloud tooling should work — full auditability, standard AWS security model, no new credential management.

Role-based orchestration (legacy). The Core MCP Server’s role system was a genuine innovation — bundling related servers into named roles. It’s now deprecated since modern MCP clients handle multi-server configurations natively, but the idea influenced how AWS organizes its server documentation.

The Serverless MCP Server is a standout. SAM init, build, deploy, local invoke, logs, metrics, plus guidance tools for architecture and Lambda best practices. Read-only by default with explicit --allow-write and --allow-sensitive-data-access flags. This is the complete serverless development lifecycle in one server.

Proactive security patching. The April 2026 releases upgraded aiohttp to 3.13.3 to resolve 8 CVEs and bumped urllib3 from 2.6.0 to 2.6.3. Dependency security is taken seriously.

What Doesn’t Work Well

Security fixes keep coming. The two CVEs from early 2026 (CVE-2026-4270 file access bypass, CVE-2026-5058 command injection) were patched, but April brought more: the IAM MCP Server had a privilege escalation vulnerability fixed (April 10), and EKS got a path traversal fix in app manifest generation (March 25). That’s four distinct security issues across four months — the pace reflects both the attack surface of 54 servers and AWS’s responsiveness, but the frequency is concerning.

EKS security bypass is still open. Issue #2942 reports that list_k8s_resources bypasses the --allow-sensitive-data-access flag, allowing enumeration of secret names, namespaces, labels, and annotations. The documented redaction logic (HIDDEN_FOR_SECURITY_REASONS) doesn’t appear to exist in the actual codebase. A PR (#2948) is pending. This is now the third EKS security issue — after the secret exposure (#2588) and the path traversal fix. The EKS MCP Server did gain kubeconfig/OIDC authentication support (April 2), but the security track record remains shaky.

Overwhelming complexity. 54 active servers is not a feature — it’s a configuration nightmare. The Core MCP Server that was supposed to simplify this with role-based groupings is deprecated. AWS’s answer is “modern MCP clients support multi-server configurations natively” — which is true, but doesn’t help with the “which of the 54 servers do I actually need?” question. Meanwhile, Google offers 50 managed remote servers that auto-enable when you enable the service, requiring zero local configuration.

147 open issues (up from 139). The CloudWatch MCP Server has a tool name that exceeds Bedrock Converse API’s 64-character limit (#3181), causing all LLM calls to fail. The AWS Knowledge MCP server has aggressive rate limiting causing startup failures (#2949). The openapi-mcp-server crashes on long API keys due to bcrypt’s 72-byte limit (#3093). With 334 open PRs, the backlog is growing faster than the team can review. When your own cloud’s MCP servers break your own cloud’s AI service, the integration story needs work.

stdio only (mostly). The Knowledge server supports Streamable HTTP, and the managed AWS MCP preview is remote, but the vast majority of servers are stdio-only. SSE support was explicitly removed in May 2025. For teams that want shared infrastructure MCP endpoints, this is limiting.

Deprecation wave accelerated. Nine servers are now deprecated — Core, CDK, Cloud Control API, Terraform, MSK, Code Documentation, Git Research, CloudWatch AppSignals, Nova Canvas, and Bedrock Data Automation. The April 21 release removed 12 deprecated directories entirely. Healthy consolidation, but nine deprecations in four months is painful for early adopters who built workflows around these servers.

Not designed for multi-tenant environments. The API server’s documentation explicitly states this. Some read-only operations can return AWS credentials or sensitive information in command outputs. For platform teams building shared developer tooling, this is a significant limitation.

Python 3.10+ and uv required. The entire suite is Python-based and requires the uv package manager. Node.js is needed for some servers. This is a heavier runtime requirement than most MCP servers, especially for teams that have standardized on different toolchains.

How It Compares

The competitive landscape has transformed since our original review. AWS no longer leads on server count — and the gap in managed experience has widened against them:

vs. Google Cloud MCP (50 managed servers): Google launched 50 managed remote MCP servers at Cloud Next ‘26, covering everything from BigQuery and Spanner to GKE, Cloud Run, Pub/Sub, Cloud Storage, and even Google Workspace (Drive, Gmail, Calendar, Chat). Every Google Cloud service is MCP-enabled by default with native IAM Deny policies, Model Armor for prompt injection defense, OTel tracing, and Cloud Audit Logs. Servers auto-enable when you enable the service — no local setup, no uv, no Python. This is the approach AWS should be worried about: same breadth, better developer experience, fully managed. Google matches AWS on count (50 vs. 54) and exceeds it on ease of use.

vs. Azure MCP (230+ tools, 45 services): Azure now ships MCP tools built directly into Visual Studio 2022 (not just VS 2026) as part of the Azure development workload — no extension required. 230+ tools across 45 Azure services accessible through GitHub Copilot Chat. The April 2026 Azure DevOps MCP update added WIQL query support. Azure bets on IDE-native MCP with massive tool consolidation; AWS bets on service-specific servers. Azure’s approach means zero configuration for VS users — a stark contrast to AWS’s 54 separate servers.

vs. Cloudflare MCP (4.5/5): Cloudflare earned the highest cloud MCP rating thanks to its Code Mode architecture that exposes 2,500+ API endpoints in just 1,000 tokens. AWS distributes similar functionality across 54 servers. Cloudflare is more elegant; AWS is more comprehensive.

vs. Docker MCP (3.5/5) for containers: AWS has three container-related servers (EKS, ECS, Finch) that operate at a higher level of abstraction — they manage cloud container services, not local Docker daemons. Different use cases: Docker MCP for local development, AWS container MCPs for cloud orchestration.

vs. Neon MCP (4/5) and Supabase MCP (4/5) for databases: AWS has dedicated servers for Aurora, DynamoDB, DocumentDB, Neptune, Keyspaces, ElastiCache, Timestream, and Redshift. More databases covered, but each server is narrower. Supabase’s single server covers database + auth + storage + edge functions in one package.

vs. Terraform MCP (HashiCorp): AWS deprecated their own Terraform MCP server in favor of the IaC server, and explicitly recommends HashiCorp’s official Terraform MCP for Terraform workflows. Healthy ecosystem separation.

The Bottom Line

The AWS MCP suite is no longer the most ambitious cloud MCP project — Google Cloud’s 50 managed remote servers and Azure’s 230+ tools make the field genuinely competitive. But AWS still has 54 active servers with deep, service-specific functionality, active weekly releases, and a managed remote endpoint in preview. The Knowledge server and API server’s security model remain well-designed. The proactive dependency patching and rapid security response show mature operations.

But the competitive gap has closed while AWS’s own challenges persist. Four security issues in four months (two CVEs, IAM privilege escalation, EKS path traversal), the still-open EKS security bypass (#2942), 147 open issues, 334 open PRs, and nine deprecated servers show that maintaining 54 servers at this pace is straining even AWS’s resources. Meanwhile, Google’s servers auto-enable with zero configuration, and Azure’s tools come built into your IDE. AWS’s stdio-first, Python-required approach feels increasingly dated in a managed-remote world.

For AWS-heavy teams, this remains essential infrastructure — the Knowledge server and managed AWS MCP preview are worth using today. For teams evaluating cloud MCP options, the choice has become genuinely interesting: Google Cloud for the best managed experience and automatic enablement, Azure for IDE-native integration and consolidated tooling, AWS for the deepest service-specific coverage. Start with the managed AWS MCP preview or the Knowledge server, then add specific servers as needs arise. Don’t try to enable everything at once.

Rating: 4 out of 5 — still the deepest cloud MCP integration with the most service-specific coverage, but the competitive landscape has transformed (Google matches on count, Azure exceeds on tooling), four security fixes in four months, an open EKS bypass, 147 open issues, 334 open PRs, and nine deprecated servers show the cost of maintaining this many servers at scale.

| Open Issues | 147 | | Open PRs | 334 |

This review was last refreshed on 2026-04-29 using Claude Opus 4.6 (Anthropic).