A collaboration between SmartBear and AWS, by Matt Bonner (SmartBear), Carl Luo (AWS), and Perminder Singh (AWS)

For decades, the software development lifecycle (SDLC) gave engineering teams a common framework. Requirements became designs. Designs became code. Teams tested, deployed, monitored, and improved it.

Building AI-powered software is different. Modern AI applications combine APIs, models, agents, retrieval systems, and orchestration layers. These components evolve independently. Their behavior is probabilistic, and their interactions are harder to predict. Failures often emerge only after deployment. Traditional testing remains essential, but alone it can’t provide confidence in AI systems. This is a core challenge of the AI-disrupted SDLC, where change outpaces validation. 

Quality tooling must evolve alongside AI adoption, wherein delivering application integrity – continuous, measurable assurance that your software just works as intended, with the governance to operate AI speed and scale – is critical. AI agents need secure access to API contracts, test results, production telemetry, and release data. With this context, agents can participate in development workflows alongside human engineers. 

This blog explains how the SmartBear MCP Server on Amazon Bedrock AgentCore can help you build AI quality workflows within this context. The SmartBear MCP Server connects SmartBear’s quality capabilities directly to AI agents through the Model Context Protocol (MCP). The server is runtime agnostic. It works with any MCP-compatible client or orchestration platform. When organizations need a scalable, enterprise-ready runtime, they can deploy the SmartBear MCP Server on Amazon Bedrock AgentCore via the AWS Marketplace listing. 

Key takeaways 

• Traditional testing on its own can’t give teams confidence in AI systems whose parts evolve independently and fail in ways that only surface after release. 

• AI agents earn a place in delivery when they can reach real quality data securely – API contracts, test results, and production signals – instead of reading it back from static dashboards. The SmartBear MCP Server exposes that quality context across the SDLC through one interface, and runs at production scale on Amazon Bedrock AgentCore. 

The SmartBear MCP Server: Your Quality Stack as Agent Context 

The Model Context Protocol (MCP) is an open standard. It defines how AI agents access external tools and services through a consistent interface. Agents interact with any MCP-compatible service using the same pattern, without custom integrations for every system. 

The SmartBear MCP Server implements this standard for SmartBear’s quality tooling. It makes testing, API governance, and production monitoring available as structured context for AI agents. This information becomes directly accessible during agent execution, rather than existing in separate dashboards. 

Through a single interface, an agent can: 

• Retrieve API contract definitions 

• Validate changes against consumers 

• Inspect test coverage 

• Analyze release history 

• Correlate production errors with recent deployments 

Agents can combine these capabilities within a single workflow. They reason across the full software delivery lifecycle. This shifts quality data from static reporting into actionable context for real-time decisions based on how your applications behave today. 

Running SmartBear MCP on Amazon Bedrock AgentCore 

Defining Amazon Bedrock AgentCore  

Amazon Bedrock AgentCore is a managed runtime for deploying AI agents at production scale. It handles session persistence, secure tool invocation, scalable compute, and operational observability.  

Think of it like a managed database service. You don’t build session management, failover, or audit logging yourself. AgentCore provides that infrastructure layer for agent workflows. 

Prerequisites 

Before deploying the SmartBear MCP Server on AgentCore, ensure you have the following: 

• An active AWS account with appropriate IAM permissions 

• Access to Amazon Bedrock AgentCore in your AWS region 

• A SmartBear account with API credentials for the products you plan to connect 

• A subscription to the SmartBear MCP Server on AWS Marketplace 

Deploying the SmartBear MCP Server on AgentCore gives your agent workflows three distinct advantages. Persistent session management means multi-step quality workflows retain state across invocations. Built-in enterprise security – IAM integration, VPC support, and audit logging – meets compliance requirements without custom code. And because performance scales, agent-invoked quality checks execute without bottlenecking your delivery pipeline. 

Deploying SmartBear MCP on AgentCore 

The SmartBear MCP Server is a container image available on the AWS Marketplace. You subscribe, deploy it into your own AWS environment, and connect it to AgentCore as your runtime. Because the MCP Server is runtime agnostic, you control where and how it runs. AgentCore is the right choice when you need production-grade infrastructure for agentic workflows. 

Enterprise Benefits of AgentCore Deployment

Deploying the SmartBear MCP server on AgentCore gives your agent workflows the following advantages: 

• Persistent session management — Multi-step quality workflows retain state across invocations. 

• Enterprise security — IAM integration, VPC support, and audit logging meet compliance requirements without custom code. 

• Scalable performance — Agent-invoked quality checks execute without bottlenecking your delivery pipeline. 

How a request flows 

An agent invokes a SmartBear capability through the MCP Server. AgentCore manages the session, handles authentication against the relevant SmartBear service, and logs the full interaction. The agent receives structured, actionable context back in a single response. 

Agent workflows operating across the SmartBear portfolio 

Every SmartBear product below exposes tools as MCP capabilities. Each section assumes you already know the product and focuses on the agent workflows it enables. 

SmartBear Swagger: API Lifecycle Context for AI Agents 

Swagger already holds your API contracts, validation rules, and documentation. As AI agents become API consumers, that accuracy becomes critical. AI agents fail or produce incorrect results when schemas drift. 

Through the SmartBear MCP Server, Swagger exposes API contracts, validation, and documentation as structured context. This includes SmartBear Swagger Contract Testing, SmartBear Swagger Studio, and SmartBear Swagger Portal. 

Agents you can run include: 

• Pre-merge contract agent – Checks a proposed API change against all known consumers and blocks the PR on a breaking change. 

• Governance agent – Reviews APIs in Swagger Studio, validates them against organizational standards, and confirms readiness for publication. 

• Documentation validation agent – Verifies APIs published in Swagger Portal are complete, consistent, and aligned with the latest contracts. 

SmartBear Reflect: Functional test maintenance without the overhead 

Reflect handles building and maintaining functional tests. Its tests operate at a higher abstraction level than selector-based automation, which makes them resilient to the surface changes AI applications undergo constantly. 

Agents you can run include: 

• Regression triage agent – Reviews Reflect results after deployment and classifies failures as genuine regressions or expected behavior changes. 

• Deployment gate agent – Blocks promotion to staging if functional coverage falls below a defined threshold. 

SmartBear QMetry: Coverage and release evidence on demand 

QMetry traces requirements through to execution records. When a model update touches dozens of downstream behaviors, you need evidence that the right things were tested. 

Agents you can run include: 

• Coverage gap agent – Compares new feature requirements against existing test cases and flags untested areas before development completes. 

• Release readiness agent – Reviews execution history across a release and returns a go/no-go recommendation with the evidence record attached. 

SmartBear BugSnag: Production signal looped into quality 

BugSnag correlates production errors, crashes, and performance events with specific releases. Real user monitoring captures latency spikes and session failures that matter most for AI-powered features. 

Agents you can run include: 

• Post-deployment review agent – Compares error rates before and after a model update and links regressions to the responsible release. 

• Root cause agent – Connects a BugSnag error spike to a recent deployment and flags the regression for a threshold-driven rollback decision. 

Cross-Tool Workflows: Connecting the Quality Lifecycle 

Individually, each SmartBear product gives AI agents access to a different layer of quality data. Together, they create a connected view of the software delivery lifecycle, delivering application integrity across the entire stack. This cross-tool workflow with SmartBear MCP Server and AgentCore occurs over four stages: design, test, deploy, and observe.  

Example: End-to-end quality flow 

A Swagger contract agent detects a breaking API change before a pull request merges. After the contract is updated, a QMetry coverage agent identifies untested behaviors and generates draft test cases into QMetry. Reflect validates the end-user experience. QMetry determines whether the release has sufficient evidence to proceed. 

After deployment, BugSnag identifies a new error pattern. It correlates the error with the release, the affected API contract, and the associated testing history. Engineers receive connected evidence that traces the issue from design through production. That connected evidence, from intent through production, is what application integrity looks like in practice. 

Amazon Bedrock AgentCore allows every MCP request to share the same managed session. The runtime handles authentication, state management, audit logging, and observability. Agents move across SmartBear tools without custom integration code. 

How to get started connecting quality across the stack 

SmartBear provides AI agents with quality context across the entire software delivery lifecycle. From API design and governance through testing, release decisions, and production monitoring, agents reason over one connected record of quality data instead of isolated tools stitched together. This unified visibility, supported by SmartBear MCP Server and AgentCore, enables team to keep pace in the AI-disrupted SDLC, using the MCP tools and agentic workflows your team and your agents already work in. 

You don’t have to adopt it all at once. Start with the quality context you already have and put it in front of your agents: 

• Subscribe to the SmartBear MCP Server on AWS Marketplace to deploy it in your own AWS environment. 

• Explore the SmartBear MCP Server overview and the capabilities it exposes. 

• Read the Amazon Bedrock AgentCore documentation for runtime setup and session management. 

Frequently asked questions 

Why isn’t traditional testing sufficient for AI-powered applications? 

Traditional testing isn’t alone sufficient for AI applications which combine models, APIs, agents, and retrieval systems that change independently and behave probabilistically, so many failures only show up after release. Traditional tests still catch known regressions, but they can’t cover behavior no one scripted for, which is why teams pair them with continuous, measurable assurance across the whole lifecycle. 

What is an MCP Server, and why would an AI agent need one for software quality? 

The Model Context Protocol (MCP) is an open standard for how AI agents reach external tools through one consistent interface. An MCP Server for quality lets an agent pull API contracts, test coverage, and production errors as structured context while it works, instead of switching between separate dashboards to find them. 

How do AI agents get secure access to test results and API contracts? 

AI agents get secure access to test results and API contracts by requesting them through the MCP Server, which authenticates against each connected service and returns structured results. Running the server on Amazon Bedrock AgentCore adds managed sessions, IAM integration, VPC support, and audit logging, so access stays governed and traceable. 

Can AI agents make release decisions on their own? 

No. In these workflows agents gather and connect evidence, such as coverage gaps, contract breaks, and error spikes tied to a release, then surface a recommendation. People still make the call, and governance and audit trails keep every action reviewable. 

What are the benefits of connecting quality data to AI agents? 

Connecting quality data to AI agents results in dewer surprises later in the lifecycle. When an agent can trace an issue from API design through testing to production in one place, teams catch breaking changes before merge, spot untested behavior before release, and link production errors back to the specific change that caused them. 

About the co-authors 

Matt Bonner is a Solutions Architect at SmartBear, where he drives technical strategy across strategic partnerships including AWS, Atlassian, and GitLab. He focuses on architecting AI-powered solutions and agentic workflows using technologies such as Amazon Bedrock, Amazon Bedrock AgentCore, and Model Context Protocol (MCP), connecting emerging AI capabilities with software development and quality engineering. Matt has spent the last seven years at SmartBear helping organizations modernize how they build, test, and deliver software.
LinkedIn – https://www.linkedin.com/in/-matthew-bonner/

Carl  Luo is a Partner Solutions Architect at AWS. He works closely with AWS Partners, diving deep into their solutions to help them integrate AWS services, build AI-powered applications, and adopt agentic architectures using Amazon Bedrock and AgentCore. Carl helps partners deliver scalable, durable, and intelligent applications to their customers. Prior to this role, he spent 18+ years as a principal engineer and staff data engineer, helping companies re-architect and migrate to the latest cloud technologies.
LinkedIn – https://www.linkedin.com/in/carl-luo-b0774958/

 
Perminder Singh is a Partner Architect and Emerging Technology Advisor at AWS, where he has worked since 2019 with ISVs, private equity portfolio companies, and startups to  bring agentic AI, physical AI, and quantum technologies from concept to production. He focuses on helping enterprise software companies architect scalable AI solutions on AWS. He has 20+ years of consulting experience at firms such as Deloitte, Capgemini, IBM, and TCS. Outside of work, he writes about the intersection of emerging technology and business transformation. 
LinkedIn - https://www.linkedin.com/in/quantumtechnocrat/