Fanroute Live

An MCP server that lets Alexa+ find the concerts you'd love, and pull the artists you want to your city.

Comment

Inspiration

Voice is how people already ask "what's on this weekend?" at home, but today's answers aren't personal. And independent artists can't prove a city will turn out before they risk booking a room. We wanted Alexa+ to answer the fan's question with their own taste, and to turn "I wish they'd come here" into something an artist can act on.

What it does

Say "Alexa, any shows I'd like this month? I love Clairo." Fanroute Live expands that taste into similar artists using real listening-session data, then matches them against the month's music events in your city from Ticketmaster, headliners and support acts. Alexa reads out the best three, and the Echo Show screen shows the rest as a card with ticket links.

Say "Alexa, I want Juniper Static to play Chicago." Fanroute Live finds the artist's open pledge campaign and tells you how close Chicago is to its goal. It puts a QR code on the screen so you can pledge $5 from your phone. The pledge is a PayPal hold, not a charge: you only pay if the Chicago show is confirmed. Fans get a way to pull a tour to their city, and independent artists see real money-backed demand before they book.

How we built it

  • MCP server over Streamable HTTP at /api/mcp, served statelessly with the MCP TypeScript SDK v2 (createMcpHandler). It speaks spec 2025-11-25 Streamable HTTP; we test it with both a v2 client and a 2025-11-25 client.
  • Four tools: find_concerts, similar_artists, bring_artist_to_city, pledge_progress. Each returns a short spoken answer plus structured data, and the concert and pledge tools return an MCP Apps view sized for an Echo Show (large type, host theme, links opened through the host). The QR code travels in _meta, so the screen gets it without filling the model's context.
  • Agent Skill: skills/fanroute-live/SKILL.md teaches any agent the voice etiquette: ask for a taste before searching, read three shows at most, never read URLs aloud, always say a pledge is a hold.
  • Data: taste from MusicBrainz + ListenBrainz session-based similarity (open data); listings from the Ticketmaster Discovery API; pledge campaigns from Fanroute, our tour-planning app, using PayPal Orders v2 authorizations that are captured only for confirmed cities and voided otherwise (sandbox in the demo).
  • Simulated Alexa+ experience: a web Echo Show with voice in and out. It connects to the MCP server as a real MCP client over Streamable HTTP and renders the views in sandboxed iframes through the official MCP Apps AppBridge. A live panel shows every tools/call and result. In the live demo, gpt-oss-120b on Groq drives the tools it discovers over MCP; with no model key, a deterministic intent router keeps the demo working.

Challenges we ran into

  • Cards under strict sandboxes. MCP Apps views run in a host-controlled iframe with a strict CSP, so we inline the MCP Apps SDK into each view at build time. Each card is one self-contained HTML document.
  • Voice-first answers. Every tool writes its own short spoken summary and keeps the details for the screen.
  • Serverless MCP. Stateless per-request servers suit Vercel functions.
  • Real Alexa+. The Alexa+ add-on path (Alexa AI CLI) requires OAuth account linking and sub-500 ms responses; our tools are public and call third-party APIs, so this version runs on a simulated Echo Show. Details are in our friction logs.

Accomplishments that we're proud of

  • A fan-to-artist loop no assistant offers today: taste-matched shows, and money-backed demand for the shows that don't exist yet.
  • Cards that render under a strict CSP, and an Agent Skill that keeps the voice experience short and honest.

What's next

  • Ship as an Alexa+ add-on with the Alexa AI CLI: pledges need fan accounts anyway, so add OAuth 2.1 account linking, and cache listings to meet the 500 ms budget.
  • "Tell me when they announce Chicago" notifications.
  • Venue-side tools for promoters.

Try it

Built With

Share this project:

Updates

Submission history