Securing AI Boundaries: Platform Teams as MCP Stewards
Platform teams can serve as MCP stewards by establishing governance frameworks that balance developer productivity with security, including creating curated registries of approved MCP servers, providing security reviews for custom implementations, and offering standardized hosting solutions while maintaining comprehensive documentation and best practices.
Two significant trends are currently reshaping software development across AWS customers. First, organizations are increasingly investing in platform engineering teams to simplify development workflows and ensure consistent governance, including providing AI coding assistants as part of their well-established tooling-as-a-service offerings. In a more nascent but transformative development, the emergence of Model Context Protocol (MCP) has enabled seamless integration between AI-assisted software development and external data sources and tools. As platform teams are uniquely positioned to provide well-architected compliant MCP servers while enabling productivity benefits, organizations must empower their platform teams to create comprehensive MCP strategies that align with broader organizational goals. These parallel trends create both opportunities and challenges that forward-thinking organizations must address.
Securing AI Boundaries: Platform Teams as MCP Stewards
Platform Engineering: Empowering developers through standardization
At its core, platform engineering treats developer experience as a product, with platform teams serving as internal product teams that build and maintain tooling-as-a-service offerings. These teams provide concrete value through self-service infrastructure provisioning, golden paths for common development patterns (such as creating new micro-services on AWS or adding observability to application components), and increasingly, AI-powered development tools like Kiro . Beyond tooling provision, successful platform teams function as enablers and advisors, offering consultation and support to development teams navigating complex technical decisions.
Model Context Protocol: Extending AI capabilities
AWS published several blogs illustrating the value of Model Context Protocol (MCP) last year. MCP is an open protocol developed by Anthropic that provides a standardized way to connect AI models to virtually any data source or tool through a client-server architecture, enabling more powerful, context-aware AI agents that can seamlessly access needed information and tools without requiring application code changes when servers are updated. For developers working with code assistants like Kiro, MCP integration delivers tremendous value by extending AI capabilities beyond built-in knowledge, allowing code assistants to access specialized domain expertise, automatically enforce best practices, and dramatically reduce research time.
MCP: A double-edged sword
Teams considering MCP adoption must carefully balance the significant productivity benefits against substantial operational challenges. While MCP servers enable powerful AI assistant capabilities by connecting to enterprise data sources and specialized tools, they introduce complex security considerations including expanded attack surfaces, credential management across multiple servers, and potential data exposure risks. These challenges are particularly evident in the broader MCP ecosystem, where the proliferation of MCP servers available on the internet presents a significant governance challenge for organizations. These servers, often easily downloadable and integrable with developer environments, can introduce security vulnerabilities if not properly vetted. Developers eager to enhance their productivity might incorporate third-party MCP servers without considering potential risks such as data exfiltration or insecure API implementations. This creates a governance gap where organizational security policies may be inadvertently bypassed. MCP servers can access sensitive corporate data, codebase information, and potentially production systems if given appropriate credentials. Without proper oversight, organizations risk exposing intellectual property, customer data, or infrastructure details to untrusted third parties. Furthermore, the quality and reliability of publicly available MCP servers vary widely. Some may contain outdated information, implement poor security practices, or lack proper maintenance. This inconsistency can lead to technical debt and security issues that compound over time as these components become embedded in development workflows. Additionally, the distributed nature of MCP architecture creates operational overhead requiring dedicated monitoring, version management, and troubleshooting expertise, while performance implications such as network latency, resource consumption, and reliability dependencies can degrade user experience. Additionally, the configuration complexity and maintenance burden may overwhelm individual developers, suggesting that successful MCP implementations require centralized platform team management with robust governance frameworks, comprehensive security controls, and standardized deployment patterns to realize the technology's benefits while mitigating its inherent risks.
Platform teams are uniquely positioned to address this governance gap, providing the necessary guardrails while still enabling the productivity benefits that MCP servers offer.
How platform teams can help
Creating an MCP registry or portal
Platform teams can establish an internal MCP registry or portal that serves as the official source of truth for approved MCP servers within the organization. This registry could be part of an existing Internal Developer Platform (IDP) maintained by the platform team. The registry should include:
- Curated collections of pre-approved MCP servers that meet security standards
- Usage metrics to understand adoption patterns
- Integration guides specific to the organization's development environments
- Feedback mechanisms for developers to report issues or request enhancements
Organizations can take different approaches to MCP governance based on their security posture and culture. Some use registries as informational resources to help developers discover vetted options, while others require enforcement mechanisms that restrict installation to approved servers only. Kiro now provides the latter capability through an MCP registry feature. Enterprise admins can specify a JSON file containing a vetted list of approved local (stdio) and remote (http) MCP servers, restricting users to installing only those servers. This registry format follows the standard introduced by Anthropic and others. This capability is currently available in Q Developer plugins , Kiro CLI , and Kiro IDE .
Developing comprehensive MCP guidance
Platform teams should create and maintain internal documentation that covers detailed information on using approved MCP servers, step-by-step guides for creating custom MCP servers, and security considerations with clear review statuses for each available server. This documentation should also include deployment models comparing local execution versus remote hosting options, best practices for data handling, authentication, and authorization, along with troubleshooting guides and common pitfalls to avoid.
Providing security reviews for in-house MCP servers
As development teams begin creating their own specialized MCP servers, platform teams can offer security review services to ensure these custom implementations adhere to organizational standards. This process should include code reviews focusing on security best practices, authentication and authorization mechanism validation, data handling and privacy compliance checks, performance and scalability assessments, and integration testing with approved development environments.
Developing and supporting core MCP servers
Platform teams can take ownership of developing and maintaining critical MCP servers that provide organization-specific capabilities. These servers would connect to the relevant company sources and knowledge bases to provide access to powerful tools and resources. Examples could be MCP servers for issue tracking systems, or collaborative productivity suites. To avoid becoming bottlenecks, platform teams should adopt an inner-sourcing model where they maintain core functionality while accepting contributions from other teams. This collaborative approach ensures that MCP servers evolve to meet changing needs while maintaining quality standards.
Hosting and infrastructure support
Platform teams can provide standardized hosting solutions for remote MCP servers. With the recent release of Amazon Bedrock AgentCore , organizations now have a purpose-built, serverless hosting environment for deploying and running MCP servers at enterprise scale, offering framework-agnostic deployment, support for any LLM, and native MCP protocol support. Amazon Bedrock AgentCore addresses operational challenges through session isolation with dedicated microVMs, consumption-based pricing, built-in authentication integrating with corporate identity providers like Okta or Microsoft Entra ID, and agent-specific observability. The service provides a unified SDK for accessing complete AgentCore capabilities including Memory, Tools, and Gateway. Building on AgentCore's foundation, Amazon Bedrock AgentCore Gateway provides an enterprise-grade solution for MCP gateway patterns at scale. Gateway transforms existing APIs, Lambda functions, and enterprise services into MCP-compatible tools through a single, secure endpoint. Critically, AgentCore Gateway now treats MCP servers as first-class integration targets alongside REST APIs and Lambda functions, enabling platform teams to unify their MCP architecture with a single point of control for routing, authentication, and tool management. The service includes built-in integrations with popular enterprise tools like Salesforce, Slack, and Jira, while providing semantic tool discovery capabilities. Additionally, AgentCore includes a Browser Tool that provides a fully managed, pre-built cloud-based browser enabling AI agents to interact seamlessly with websites. When combined with Amazon Nova Act (preview)—AWS's browser automation SDK—this capability bridges the gap for scenarios where REST APIs aren't available, allowing agents to navigate web interfaces and complete tasks like form submissions that would otherwise require manual intervention or custom automation scripts. With general availability announced in October 2025, all AgentCore services now support Virtual Private Cloud (VPC) connectivity, AWS PrivateLink, AWS CloudFormation, and resource tagging, enabling developers to deploy AI agents with enhanced enterprise security and infrastructure automation capabilities critical for production deployments. For teams that want to run their own MCP servers, platform teams can provide blueprints according to organizational best practices, whether using AgentCore for managed hosting, AgentCore Gateway for enterprise tool integration and MCP server unification, or self-managed infrastructure approaches using AWS Lambda or ECS Fargate. For organizations preferring open-source solutions, community projects like MCP Gateway & Registry provide alternative implementations for centralized MCP management and discovery.
Measuring success
When evaluating the effectiveness of your MCP governance program, platform teams should leverage the same metrics traditionally used to measure developer experience and platform engineering success. These include developer velocity gains (commit-to-production time, deployment frequency), platform adoption rates, net promoter scores from internal teams, and direct business impacts such as reduced time-to-market for new features. Additionally, track MCP-specific metrics like the percentage of approved versus unapproved server usage, security vulnerabilities prevented through the review process, and knowledge reuse across teams.
Conclusion: enabling innovation with guardrails
The convergence of platform engineering and MCP represents a significant opportunity for organizations to accelerate development while maintaining appropriate governance. By establishing clear governance models for MCP servers, platform teams can strike the right balance between innovation and security. This approach allows organizations to fully realize the productivity benefits of AI-assisted development while mitigating potential risks. If your organization is exploring or already implementing MCP servers, now is the time to involve your platform engineering team in creating a governance strategy. A critical pitfall to avoid is creating excessive bureaucracy that undermines developer autonomy—ensure your MCP governance process remains lightweight, automated where possible, and focused on enabling rather than controlling development teams.
Additional Reading
About the Authors
Mohammed Issa is a Principal Customer Solutions Manager at AWS based in Sweden. Bringing long expertise in AWS cloud services, developer experience, DevOps, AI/ML and enterprise transformation, he helps customers adopt complex modern technologies and drive business value from their technology investments. Outside work, he is a husband and a father of 2 sons, and enjoys movies, football, and hanging out with friends.
Bastian Töpfer s a Senior Solutions Architect at AWS with over 20 years of experience developing and architecting software solutions for enterprises and start-ups across telecom, healthcare, automotive, and social media. He works with customers to design, implement, and optimize cloud-based solutions, leveraging expertise in infrastructure, software development, and machine learning to deliver scalable, secure, and cost-effective outcomes. Bastian is passionate about learning new technologies and sharing his knowledge with others.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article