AIToday
Large Language ModelsAI Coding AssistantsAmazon AI BlogPublished: Sep 12, 2026, 04:00 JST2 min read

AWS shows MCP Apps on Amazon Bedrock AgentCore

AWS shows MCP Apps on Amazon Bedrock AgentCore

3 Key Points

  1. What happened

    AWS published a step-by-step demo of Unicorn Rentals, an MCP App that lists, books, and returns unicorns, running on Amazon Bedrock AgentCore runtime and exposed through AgentCore Gateway. The same server renders the same interactive cards in ChatGPT and Claude.

  2. Why it matters

    MCP Apps is host-agnostic, so organizations reach customers on ChatGPT, Claude, or other supported hosts without rebuilding for each. The thin protocol adapter delegates business logic to AWS Lambda and DynamoDB, which could be existing services.

  3. What to watch

    Since AgentCore runtime uses consumption-based pricing, production success hinges on right-sizing container memory and reviewing Lambda concurrency. AWS recommends AWS WAF, CloudWatch alarms, and Amazon Bedrock Guardrails at trust boundaries.

WHO IT HITSEnterprise developers and platform teams building customer-facing AI integrations can use this pattern to add interactive widgets without coupling to a single AI host, while keeping their existing business logic on services like AWS Lambda, Amazon ECS, or Amazon EKS.

Ask the AI about this article →

Summaries like this, in your inbox every morning.

Context & Analysis

The post arrives as customers increasingly interact with digital services through AI hosts like ChatGPT and Claude, which creates a need to serve rich UI across those applications without tying the service to any single host. AWS positions MCP Apps, an extension of the Model Context Protocol, and Amazon Bedrock AgentCore as the pairing that meets both needs. The walkthrough uses Unicorn Rentals to show four steps: list available unicorns as cards, book one by name, view active bookings as text, and return the unicorn with the total cost calculated from duration and hourly rate.

Architecturally, the MCP app is a thin protocol adapter that delegates business logic to an AWS Lambda function backed by Amazon DynamoDB. AgentCore runtime hosts the app in a secure, serverless, session-isolated environment with native MCP support, while AgentCore Gateway exposes it through a single managed endpoint. AWS notes the Lambda could represent existing services on Amazon ECS, Amazon EKS, or other compute, which suggests a path for organizations that already run business logic on AWS. The same server is shown connecting to ChatGPT Developer Mode and Claude Connectors, and AWS states it works with other MCP Apps-compatible hosts.

Looking ahead, the approach's value for teams may hinge on how much operational work AgentCore removes, since AWS says runtime handles scaling, session isolation, health monitoring, and infrastructure management. The production cautions around consumption-based pricing, CloudWatch monitoring, and Amazon Bedrock Guardrails are likely where the real trade-offs surface for anyone moving beyond the sample.

FAQ
Do I need to rebuild the app for each AI host?
No. Because MCP Apps is a host-agnostic standard, the same server delivers the same rich experience across AI hosts that support the extension, including ChatGPT and Claude.
What does the Unicorn Rentals demo actually do?
Users can browse available unicorns as interactive cards, book a unicorn by name, view active bookings, and return a unicorn. Booking confirmations show a booking ID, date, hourly rate, and the unicorn booked.
How is the MCP App secured and monitored?
The AgentCore Gateway endpoint uses AWS WAF for IP allowlisting, threat detection, and rate limiting. For production, AWS recommends Amazon CloudWatch alarms for error rates and latency, plus Amazon Bedrock Guardrails to filter harmful content and redact PII.
Amazon AI BlogRead Original Article

Get the latest Large Language Models news every morning

For example, today's edition would include:

  • Dynatrace acquires Arize AI as observability shifts to actionSiliconANGLE AI · 4h ago
  • Shared base cuts 100 fine-tunes from 1.5 TB to 19.3 GBDaily Dose of Data Science · 4h ago
  • OpenAI agents hit RubyGems, undisclosed since May 12thSimon Willison's Weblog · 4h ago

AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. Q&As are published on this page for other readers too.

Related Articles

Next articleOpenAI's Habitat storage rewritten in Rust, 6x more CPU efficient