How to enter Amazon's $190K hackathon with an MCP server you already built

Here's a strange fact about the current AI landscape: if you spent any part of this year building a Model Context Protocol server for Claude or another AI host, you're not starting from zero on an Alexa+ integration. You're most of the way there already. Amazon didn't invent a new proprietary format for outside developers to learn — it adopted MCP as the standard for how services plug into Alexa+, which means your existing tool definitions, your existing server code, and your existing understanding of how MCP works all transfer over with minimal rework.

And Amazon just put real money behind that compatibility. The company's global developer hackathon, Build, Ship, Shape, is offering a combined $190K in cash and AWS credits across four tracks, and the Alexa+ track is built specifically around MCP servers and Agent Skills.

This guide walks through what the hackathon actually wants, how the Alexa+ MCP integration works technically, and how to get a working submission together without needing a physical Echo device on your desk.

What is the Build, Ship, Shape hackathon?

Build, Ship, Shape is Amazon's developer hackathon covering four separate tracks: Alexa+, Fire TV, Ring, and Bee (Amazon's wearable AI device). The official Devpost page frames it as an opportunity to connect Fire TV, Alexa+, Ring, and Bee in a single project and compete for $190K in cash and AWS credits, with category winners also getting a 1:1 strategy session with Amazon Developer leadership.

You pick a primary track for your submission. According to the official hackathon rules on Devpost, you pick a primary track and ship a working demo, and you can layer a mini challenge onto the same project for extra ways to win — though each project can only win one track prize and one mini challenge prize.

Submissions are open now. Submissions close October 23, so there's a real but manageable runway to get something built and demoed.

Why does the Alexa+ track matter so much for MCP developers?

This is the track built explicitly for people who already work with agentic tools and MCP. Per Devpost's track description, the Alexa+ challenge asks developers to build a self-hosted MCP server (spec 2025-11-25 or later, Streamable HTTP) or an Agent Skill, showing how a brand or service could plug into the assistant.

Amazon is explicit that MCP is the backbone of this integration path. The docs state plainly that MCP is the open standard that powers Alexa+ integrations. If you've never touched MCP before, the hackathon organizers don't leave you stranded — they point new builders toward existing scaffolding tools, and the resources page notes that even without an Echo you can build a simulated Alexa+ experience in a web app using your preferred agentic tool.

That last point matters more than it might seem. A lot of people assume a hackathon centered on a voice assistant requires physical hardware to test on. It doesn't. You can build and demo the entire flow — MCP server exposing tools, Alexa+ (or a stand-in for it) calling those tools, results coming back — inside a browser-based simulation.

How does the Alexa+ MCP integration actually work?

Under the hood, Alexa+ behaves like any other MCP client. Amazon's own developer documentation lays out the architecture clearly: the Model Context Protocol is an open standard that lets AI assistants discover and use external tools and services, and with the Alexa+ MCP Toolkit, you connect your MCP server to Alexa+, giving customers access to your capabilities through voice and visual interactions on Alexa-enabled devices.

You don't write anything Alexa-specific to expose functionality. You write a normal MCP server. Amazon's documentation confirms this directly: you can bring your existing MCP server built for other AI hosts, or build a new MCP server that exposes your service's capabilities — Alexa+ acts as the MCP client, and at runtime, when a customer makes a request intended for your service, Alexa+ routes the request through your add-on to your MCP server using streamable HTTP. Alexa+ then invokes the appropriate tools and returns the result back to the customer.

If you ever update your tool definitions, you don't need to resubmit anything to an app store queue for basic changes — you redeploy your add-on with alexa-ai deploy, and your new tools and other changes take effect after redeployment.

The full technical breakdown lives in the Alexa+ MCP Toolkit Overview, which is the primary doc you should bookmark before writing a line of code.

What are the technical requirements for a submission?

There are a handful of hard requirements worth locking in before you start building, straight from the Alexa+ MCP QuickStart Guide:

Transport: your MCP server must support Streamable HTTP — the MCP specification (2025-11-25) deprecated the standalone HTTP+SSE transport in favor of Streamable HTTP, so if your MCP server currently uses the legacy SSE transport, you must migrate before connecting to Alexa+. If you built your server sometime in the last several months on a current SDK, you're likely already compliant.
Public accessibility: your MCP server must be accessible via a remote URL, and during local development, you can use a tunneling service (such as cloudflared) to expose your server. This is a five-minute setup, not a deployment headache.
Auth handling: your MCP server returns 401 Unauthorized without a WWW-Authenticate header for unauthenticated requests, hosts a Protected Resource Metadata document at the well-known URI according to RFC 9728, and makes your auth server metadata available at /.well-known/oauth-authorization-server. If you're not planning on gated access for your hackathon demo, you can often keep this minimal, but it's worth reading the full auth section before you submit.
Visual UI (optional): if your integration needs a screen component rather than pure voice/text, follow the MCP Apps standard — the MCP Apps SDK (@modelcontextprotocol/ext-apps) provides Agent Skills that guide AI coding agents to scaffold a new MCP App with an interactive UI from scratch, or convert an existing OpenAI App to use MCP Apps.

Getting onboarded is handled through Amazon's own CLI. The quickstart notes you can onboard your MCP server to Alexa+ using the Alexa AI CLI, and if you use an AI coding agent (Claude Code, Kiro, Codex, Cursor), install the Add-on Agent Skill for faster onboarding. That's a nice touch — the same agentic coding tools you already use to build the server can also handle the Alexa+ onboarding steps for you.

Authentication with Amazon itself is straightforward too: after you install the CLI, you authenticate with your Alexa developer account, which opens a browser for Login with Amazon (LWA) OAuth, and credentials get stored locally for future CLI use.

What does the prize structure actually look like?

Beyond the general Devpost framing of category winners getting a 1:1 strategy session with Amazon Developer leadership on top of cash and credits, the Alexa+ track specifically carries a $25,000 cash prize plus $15,000 in AWS credits for first place, based on the figures Amazon has publicized around the hackathon. Community trackers following the hackathon have echoed the same split across cash and credit tiers for the top spots.

While you're building, you don't have to eat the AWS bill yourself. The resources page notes you can request $150 in AWS credits through the credit request form just for participating, which is enough to cover a Lambda-hosted MCP server, a small database, and some inference calls for a demo-scale project.

What should you actually build for the demo?

Amazon's own framing gives you a hint about what they're prioritizing. The Alexa+ track description on the resources page points builders toward real, working integrations rather than toy demos — the emphasis throughout is on shipping something functional, not conceptual.

A few practical angles that map well to an MCP-first submission:

Take an MCP server you've already built for Claude Desktop or another host and port it. Since Alexa+ acts as a standard MCP client, most of your tool logic doesn't need to change — you're mainly handling transport and auth requirements.
Build something genuinely voice-native. Tools that make sense to trigger by talking — checking a status, kicking off a task, getting a summary — play to Alexa+'s strengths better than tools that assume a screen.
Simulate the full loop in a web app if you don't have Echo hardware. Since the whole point of MCP is a standardized request/response contract, you can stand up a lightweight client that mimics how Alexa+ would call your server, and demo that instead.
Document your MCP tool schemas clearly. Judges reviewing dozens of submissions will move faster through a server with clean, well-named tools and obvious descriptions than one with vague or overlapping capabilities.

How do you actually enter?

Submissions run through Devpost. Head to the Build, Ship, Shape hackathon page to register, read the full track-by-track rules, and start a project page. Amazon's resource hub also links out to track-specific documentation, including the Alexa+ Developer Docs home, which is worth skimming even if you only plan to build for one track — the "Set up," "Create an MCP Add-on," and "Client lifecycle" pages cover everything from your first deploy to handling session state.

You don't need a team, prior Amazon developer experience, or specialized hardware to get in the door. What you need is a working MCP server, a clear demo of it doing something useful, and a submission before the October 23 deadline.

If you've been sitting on an MCP server that only ever ran locally against Claude Desktop, this is about as low-friction as a hackathon gets — you're not learning a new protocol, you're pointing a protocol you already know at a new, much larger audience. Millions of people who'll never open a terminal are about to start talking to tools built exactly like the one collecting dust in your repo. Might as well be yours they talk to first.