briefings

briefing-mcp

MCP server for Briefing Service: every hour an LLM editor ranks ~100 feeds per topic (AI news, AI research, frontier labs, markets, US sports, European football, US news, world, crypto & onchain) plus two daily editions (Grants & Funding: newly posted federal funding notices with Deadline / Amount / Eligible on every item; Noticias Lentas: the day's news in simple Spanish at CEFR B1 with a glossary) and publishes the result as structured JSON, a rendered 480x800 e-ink page, a multi-page Morning Paper PDF, and a free provenance dump of every candidate it saw.

This package runs locally over stdio and calls the public REST API. The same tools are served remotely (streamable HTTP, no install) at https://briefing-service.wholemind.workers.dev/mcp.

Install

Claude Desktop / Claude Code / Cursor / any stdio client:

{
  "mcpServers": {
    "briefings": {
      "command": "npx",
      "args": ["-y", "briefing-mcp"],
      "env": { "BRIEFING_KEY": "" }
    }
  }
}

Claude Code one-liner:

claude mcp add briefings -- npx -y briefing-mcp

Claude Desktop, one click: download briefing-mcp-0.1.0.mcpb and open it (MCP bundle; it asks for the optional Reader key).

Tools

tool	what it returns	cost
list_briefings	the 11 keys (ai, research, labs, finance, sports, soccer, us, world, crypto, grants, slow-es-b1), names and prices	free
get_briefing(key)	lead, ranked stories, research, why-it-matters, key points	25 free/IP/day, then key or x402
get_summary(key)	headlines and one-liners	free
get_candidates(key)	every candidate this hour: canonical link, outlets, selected, collapsed duplicates with duplicate_of	free
get_page(key, page, wide)	480x800 1-bit PNG (or 800x480 wide front page)	25 free/IP/day

Pricing

• Free: 25 JSON/page/tool calls per IP per day, plus the summary and candidates endpoints.
• Reader key: $9/month, unlimited reads and Morning Paper delivery. Set it as BRIEFING_KEY. Buy at https://briefing-service.wholemind.workers.dev/billing/checkout
• Per call: x402 (USDC on Base) against the remote endpoint.

Docs for agents: https://briefing-service.wholemind.workers.dev/llms.txt

How it works

index.js is the whole server: it registers five tools with @modelcontextprotocol/sdk and answers each by calling the public Briefing Service REST API over HTTPS (GET /v1/briefings/...). It stores nothing, runs no shell commands, and reads only two environment variables (BRIEFING_KEY, BRIEFING_BASE_URL). The Dockerfile builds the same thing; docker run -i speaks MCP over stdio.

Run the smoke test locally:

npm install
node client-test.mjs   # lists tools and calls list_briefings, get_summary, get_candidates, get_page, get_briefing

Write-up of the pipeline (collect, entity-overlap dedupe, LLM editor, 1-bit render) and what the first week taught: https://dev.to/jshelley/i-built-an-hourly-newspaper-for-e-ink-and-turned-the-pipeline-into-an-mcp-server-1ebg. Each briefing also has an Atom feed at /v1/briefings/{key}/feed.xml; the Morning Paper PDF carries an outline per story, front-page go-to links and a clickable source URL on every story page.

Environment

• BRIEFING_KEY: optional Reader key (sent as Authorization: Bearer).
• BRIEFING_BASE_URL: override the API base (default: the public service).

License

MIT