Skip to main content
Glama

Evening Briefing — an Alexa+ MCP server

One self-hosted MCP server that turns any MCP-capable assistant into an evening companion: "Alexa, how's it looking outside, who won, and what's on tonight?" Three voice-first tools, zero API keys, zero accounts — clone, npm install, npm start.

Built for the Build, Ship, Shape: Amazon Developer Hackathon — Alexa+ track (self-hosted MCP server, Streamable HTTP, official @modelcontextprotocol/sdk).

Run it

npm install
npm start          # → http://localhost:8787
  • Playground (simulated Alexa+ experience): open http://localhost:8787 — three cards, real MCP calls over POST /mcp.

  • Any MCP client (Claude Code, Inspector, …):

claude mcp add evening-briefing --transport http http://localhost:8787/mcp
# or: npx @modelcontextprotocol/inspector
npm test           # boots the server, connects a real MCP client, calls all three tools

Related MCP server: Multi-MCPs

Tools

Tool

Source (keyless)

Answers

get_weather

open-meteo

"It's 22°C and mostly clear in Istanbul — rain chance 0%, no umbrella needed."

get_scores

ESPN public API

"Final: Celtics 112 – Knicks 104." (nba, nfl, mlb, nhl, epl, ucl)

get_tv_schedule

TVMaze

"Tonight: 19:00 The Weeknight (MS NOW), 20:00 …, 9 more."

Every tool returns a speakable one-line summary (content) plus structured data (structuredContent) — designed for voice output first, screen second.

Architecture

  • src/server.ts — MCP server over Streamable HTTP (stateless mode), also serves the playground. Plain node:http, no web framework.

  • src/tools.ts — the three tools; small, cache-free, timeout-guarded fetch wrappers.

  • public/ — vanilla-JS playground that speaks raw JSON-RPC to /mcp (initialize → tools/list → tools/call).

  • test/smoke.ts — integration test: official MCP client, ephemeral port, live tool calls.

Product feedback (hackathon requirement)

  • What we built on: the Alexa+ MCP story is the strongest part of the stack — pointing a self-hosted server at a living-room assistant is a genuinely novel distribution channel, and Streamable HTTP made it a one-afternoon integration with the official SDK.

  • Friction: discovery of which spec version / transport a given assistant firmware accepts is opaque; a public "skill ↔ MCP endpoint" debugging console (like the Inspector, but assistant-side) would save every participant a day.

  • Request: per-tool OAuth scopes and a revocation UI. Home data is intimate; trust tooling will matter more than tool count for adoption.

  • Most-wanted next tools: calendar read access and Echo device presence — the briefing becomes truly personal once the assistant knows who's home.

Track fit & scope

  • Track: Alexa+ (self-hosted MCP server path). No Amazon hardware or account needed to run or judge this.

  • Stretch (not required): the same tools deploy unchanged behind Bedrock AgentCore for the AWS Builder mini-challenge — kept out of scope here to stay keyless and instantly runnable.

  • License: MIT.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      C
      quality
      D
      maintenance
      Aggregates multiple third-party APIs into unified MCP tools, providing out-of-the-box access to 10 popular services including OpenWeather, Google Maps, GitHub, Notion, Spotify, and more. Enables users to interact with weather data, search places, manage repositories, create content, and access various web services through a single MCP server.
      34
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Provides weather data including current conditions, forecasts, and summaries via Open-Meteo with no API key required, enabling natural language queries through an MCP interface.
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Exposes MCP tools over Streamable HTTP that turn ranked Sygnal signals into voice-ready buyer-intent briefings with analyst verdicts, card and carousel payloads, and session state for a simulated Alexa+ experience. It also surfaces the next recommended action so a voice assistant can guide users through follow-ups, with fallback fixtures when the live signal feed is restricted.
      MIT