Horizon: AI News & AI Regulation Tracker

sh.alchemylab.horizon由 horizon.alchemylab.sh 發布官方發布者v1.2.1更新於 Nov 5, 2026

Free, no-key AI news + AI regulation tracker MCP: EU AI Act, US federal & state, UK. Daily briefing.

已驗證Streamable HTTP可網頁執行網頁搜尋與抓取知識與記憶

概覽

AI 產生的概覽

免費、免金鑰的遠端 MCP 伺服器,讓助理取得每日 AI 新聞簡報、AI 監管動態,以及附證據標籤的地區與主題訊號。

可以做什麼
  • 取得今日或任一過去日期的 AI 每日簡報,含來源與佐證標籤
  • 追蹤歐盟、美國聯邦與州、英國的 AI 立法,並顯示每項所處的立法階段
  • 依地區(中國、韓國、日本、歐盟)或主題取得依熱度排序的訊號,並查看報導盲區
  • 查看某公司或模型的討論中,自述與獨立評估各占多少
  • 搜尋近兩週新聞標題、依游標增量取得新項目、尋找語意相似的項目
  • 列出語料來源與篩選方式
適用情境
適合讓助理回答當前 AI 產業新聞、美國、英國或歐盟 AI 監管進展,以及地區媒體關注但西方資訊流遺漏的內容,且不想設定 API 金鑰時使用。
執行需求
需要一個遠端 streamable HTTP 端點,無需驗證、金鑰或 OAuth。僅支援 stdio 的用戶端可透過 npx 執行 mcp-remote 橋接。需要能連線至該託管服務的網路。
安裝前請注意
託管伺服器為閉源,所有查詢都會送往第三方。部分工具(附引用的問答與趨勢)需要金鑰,不在免費工具集內。結果是彙整、依熱度排序的情報,並非已查證的事實;伺服器明確表示不會判定任何內容為真或假。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Horizon: AI News & AI Regulation Tracker,將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Web executable,透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

交給代理程式安裝

把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式,或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南,先向你展示發布者、端點和登入方式,經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。

閱讀 https://sourceweft.com/mcp/sh.alchemylab.horizon%2Fbriefing/install.md,並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前,先告訴我它的發布者、端點和登入方式,等我確認。不要把 API 金鑰寫進命令或我會提交的檔案;問我金鑰應從哪裡取得。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "horizon-ai-news-ai-regulation-tracker": {
      "type": "http",
      "url": "https://horizon.alchemylab.sh/api/mcp"
    }
  }
}

README

Horizon — AI news & AI regulation tracker for agents (MCP)

A free, no-key MCP server for AI news and AI regulation tracking — the EU AI Act, US federal and state bills, UK policy — plus a daily AI briefing, all with the evidence behind each claim labelled: self-reported by the company, or independently reported and evaluated.

This repository holds the docs, the plugin and the listing files. The hosted server itself is closed source.

Claude Code plugin

Adds the MCP server and three skills — /ai-news, /ai-regulation, /ai-receipts — that teach Claude which tool answers which question and how to relay the evidence labels instead of flattening them into headlines.

/plugin marketplace add Alchemy-Lab-Infra/horizon-mcp/plugin install horizon@alchemylab

Then ask: "What's the state of the EU AI Act?" — or /ai-regulation, /ai-news, /ai-receipts directly.

Any other MCP host

Point it at the endpoint. No key, no OAuth:

json
{ "mcpServers": { "horizon": { "url": "https://horizon.alchemylab.sh/api/mcp" } } }

For hosts that only speak stdio, npx mcp-remote https://horizon.alchemylab.sh/api/mcp bridges it.

Gemini CLI: gemini extensions install https://github.com/Alchemy-Lab-Infra/horizon-mcp

VS Code and Copilot: add it to .vscode/mcp.json:

json
{ "servers": { "horizon": { "type": "http", "url": "https://horizon.alchemylab.sh/api/mcp" } } }

plugins/horizon is also a portable Agent Plugin (plugin.json, mcp.json, skills/), for hosts that read that format.

Tools

ToolWhat it answers
get_daily_briefingWhat happened in AI today, or on any past date (public archive)
get_regulation_updatesCurrent AI legislation with its stage — eu, us_federal, us_state, uk
get_entity_provenanceHow much of the talk about a company or model is self-reported vs independently evaluated
get_topic_signalRecent signal on models, regulation, research, funding, tools…
get_region_signal / get_china_signalWhat Chinese, Korean, Japanese or EU press is covering
get_blind_spotsWhat regional press covers that Western feeds miss
search_newsQuick relevance check, top 3
get_new_since / get_relatedIncremental polling and "more like this"
list_sourcesWhat the corpus covers

Key-gated: ask_horizon (cited answers) and get_trends — see /developers.

What makes it different

Most AI-news tools give you headlines. Horizon labels who is saying it: a lab's own benchmark number carries "self-reported, no independent confirmation"; a number measured by LMArena, Epoch, Artificial Analysis or METR carries "independently evaluated". Regulation items carry their legislative stage from the source, not from coverage. Nothing is ever marked true or false — the shape of the evidence is the product.

Sources: Hacker News, Lobsters, Bluesky, curated lab and media RSS, arXiv, Hugging Face, ModelScope, US Congress and federal agencies, US state legislatures, UK Parliament, EU bodies. Regional lenses for China, Korea, Japan and the EU.

License

MIT for this repository. Attribution requested: name "Horizon" and link https://horizon.alchemylab.sh when you use results; every result carries a ready-to-paste provider.citation.

來源:README.md,提交 a215919

工具

11

get_daily_briefing

Daily AI briefing

Today's AI briefing, or any past day's: top stories with their sources and corroboration labels, emerging signals, new entrants, and regulation updates. Drawn from Hacker News, Bluesky, arXiv, Hugging Face, lab blogs, tech press in the US, Europe and East Asia, and legislative sources. A curated once-daily synthesis — for raw item-level updates since your last check, use get_new_since instead. Pass a date (YYYY-MM-DD) for a past day; set top_stories_only for a shorter result.

輸入參數datetop_stories_only

get_regulation_updates

AI regulation updates (US/UK/EU)

Current AI legislation headlines across the US Congress and federal agencies, US state legislatures, the UK Parliament, and the EU, with each item's legislative stage (proposed / committee / floor action / passed / implementation). State bills are the largest share by far; filter with jurisdiction to narrow.

輸入參數limitjurisdiction

get_china_signal

China & East-Asia AI signal

Shortcut for get_region_signal with region "china": the current China & East-Asia AI signal (Qwen, DeepSeek, Zhipu, Moonshot, ModelScope, and more) that Western feeds often miss — ranked by momentum. Prefer get_region_signal if you may query more than one region.

輸入參數limit

get_region_signal

Regional AI signal

The current AI signal for a region (china, korea, japan, or eu) — recent, relevance-scored items on that region's models, labs, and analysis, ranked by momentum. Includes local-language press translated into English. The canonical regional tool; get_china_signal is a preset of this with region "china". Returns titles, sources, and links.

輸入參數limitregion必填

get_entity_provenance

Claim provenance for an entity

For one company, model, or person: how much of the claim activity Horizon holds provenance data for is self-reported by that entity versus independently evaluated by someone else. Descriptive counts over the last 30 days — not a score, and not a judgement about whether any claim is true. Answers "is the buzz around X coming from X?".

輸入參數entity必填

get_blind_spots

Coverage blind spots

Coverage blind spots: companies, models, and people that regional press (Chinese/East-Asian, European, Latin-American) is covering while Western/global feeds are not — ranked, with per-region item counts. Answers "what is Western tech media missing right now?". Measured within Horizon's corpus: "invisible" means zero pickup in the Western feeds Horizon tracks, not a claim about all media.

輸入參數dayslimit

get_topic_signal

Topic AI signal

The current AI signal for a topic — recent, relevance-scored items tagged with that topic, ranked by momentum. Like get_region_signal but sliced by subject instead of geography.

輸入參數limittopic必填

search_news

Search AI news

Quick title search over the last 14 days of AI news: returns the top 3 matches by momentum, with links and item ids (pass one to get_related).

輸入參數query必填

get_new_since

New AI items since cursor

Incremental poll: raw item-level AI news added since a cursor, oldest→newest, with a nextCursor for your next call — use this for "what's new since I last checked"; for the curated once-daily synthesis use get_daily_briefing. Omit cursor for the latest items plus a cursor to start polling from. Returns titles, links and topics.

輸入參數limitcursor

list_sources

Coverage & sources

Describe what Horizon covers: ingestion sources with their source= filter values, the regional lenses (region= / dedicated tools), and how to filter. Call this to understand the corpus before querying.

無需輸入參數。

版本歷史

3
  1. v1.2.1最新Nov 5, 2026
  2. v1.2.0Oct 24, 2026
  3. v1.1.0Sep 16, 2026

評分

已滿足適用檢查項的 100%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通必需25 分
  • 已宣告認證方式必需10 分
  • 已列出工具必需15 分
  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分