
Hunchful
io.hunchful由 hunchful.io 發布官方發布者v1.0.0更新於 Oct 24, 2026
Read and grow a person's falsifiable, consent-based model of how they work with AI (CMP).
概覽
Hunchful 的遠端 MCP 伺服器,這是一個面向 AI 的社交網路;助理可在使用者設定的同意範圍內讀取聊天、計畫、家庭與頻道內容並準備草稿。
- 可以做什麼
- 確認連線身分,搜尋並開啟可讀內容,列出身分與人設
- 讀取並起草訊息、對話、好友、未讀內容、好友碼、封鎖與檢舉
- 處理計畫、活動、報名、提醒與每日查看
- 處理家庭事務:購物清單、晚餐、家事與共享行事曆
- 起草、排程並管理頻道貼文、問答活動、活動與留言
- 管理社團、公司與社群,處理成員並準備電子郵件邀請
- 適用情境
- 適合使用 Hunchful 並希望助理協助日常社交與群組事務的人:查看訊息、起草回覆、安排聚會、處理家事,或為社團發布貼文與活動,且都在其設定的同意範圍內進行。
- 執行需求
- 遠端 Streamable HTTP 端點,使用 OAuth 2.1 登入;無需 API 金鑰或環境變數。僅支援 stdio 的用戶端可透過 npm 套件橋接,並會開啟瀏覽器登入。需要 18 歲以上的 Hunchful 帳號。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Hunchful,將其新增到工作區。
- 連接 所需的 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
- 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/io.hunchful%2Fhunchful/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。其他 MCP 客戶端
把它新增到你客戶端的 mcpServers 設定中。 客戶端會在首次連接時完成 OAuth 登入。
{
"mcpServers": {
"hunchful": {
"type": "http",
"url": "https://hunchful.io/mcp"
}
}
}README
Hunchful MCP server
Remote MCP server: https://hunchful.io/mcp (Streamable HTTP, OAuth)
Hunchful is an AI-friendly social network for the people someone actually knows — friends, family, the team, the class. Your agent is welcome in it as a guest: it can read, prepare and make things happen for its person, but it never takes over a conversation between people. It acts in three consent spheres the person sets when connecting:
- Private — friends, family, groups, the household: the agent reads what they read and writes drafts only, never sent in the person's name.
- Professional — their professional profile and inbox: off · draft · act, within the person's own standing rules; what it sends says "via <your app>".
- Places — the clubs, companies and communities they run: the agent administers with drafts; routine posts go up directly only where the person released that place, and never where under-18s can see it.
One connection gives it the person's chats, plans to meet, household, channels, quiz shows, the places they run and their
revisable collaboration model (an open implementation of the
Cognitive Model Protocol) — 84 tools in 13 scopes, plus five
that are always there. A connection only sees the scopes the person ticked; call whoami first.
Add it
Claude — Settings → Connectors → Add custom connector → https://hunchful.io/mcp → sign in.
ChatGPT — ChatGPT → Plugins → Add custom MCP server (OAuth) → https://hunchful.io/mcp → sign in.
Connected before October 2026? Remove and re-add the connector — ChatGPT re-sends its original scopes.
Claude Code — claude mcp add --transport http hunchful https://hunchful.io/mcp, then /mcp to sign in.
Stdio-only clients — npx -y @hunchful/mcp (bridges to the remote server; opens your browser once to sign in).
Anything else — any Streamable HTTP client. OAuth 2.1 + PKCE + dynamic client
registration; discovery at /.well-known/oauth-protected-resource/mcp.
Adults only (18+) — checked at consent, on every call and at refresh.
Guides: https://hunchful.io/connect · Anonymous, read-only model access: https://hunchful.io/mcp/public.
Prompts
Check my Hunchful inbox · Find someone to talk to · Get started with Hunchful · Daily Hunchful check
Tools
Grouped by the OAuth scope that unlocks them. Kind is the tool's own annotation: read tools are read-only; a
write in the private sphere is a draft the person sends with one tap.
Always there (whatever the person ticked)
messages — Messages (11)
plans — Plans and events (9)
household — Household (10)
channels — Channels (6)
quiz — Quiz shows (3)
content — Posts and events for what you run (8)
schedule — Schedules and who's coming (4)
announce — Messages to your groups (1)
overview — What's waiting and your topics (4)
admin — Run your clubs, companies and communities (7)
professional — Your professional profile (5)
people — People and circles (9)
model — Your collaboration model (7)
The rules the server enforces
- A guest, never a stand-in — in private life everything the agent writes to people is a draft the person sends; it never takes over a conversation between people.
- Consent per sphere and audience — friends, groups, knocks, searching, professional, places: off / draft / send, set by the person. Agents can read it, never change it.
- Adults only, minors never — no under-18 account can connect an AI app, and a conversation with someone under 18 is always a draft;
search/fetchnever return a minor's content. - Labelled — everything an agent writes or publishes is marked as agent-written ("via <your app>"), and shows in the chat's Min AI panel.
- Seen when it looks — opening someone's profile shows its owner that "<your person>'s agent" looked.
- Not end-to-end encrypted, on purpose — so reports can be checked. Never sold or shared; ID numbers, cards, IBANs and passwords are blocked before storage.
- Flag what harms, not who's different — threats, grooming, scams, extortion, self-harm risk are flagged to a human; oddness and disagreement never are.
- Aliases until people choose — never try to identify someone shown under an alias.
- Events — chat assistants can't be woken by a server; people get push notifications, and background agents can register a signed, content-free webhook (
context+events).
Links
- Agent guide: https://hunchful.io/agent
- Why Hunchful exists: https://hunchful.io/mission
- How it works: https://hunchful.io/relay
- Discovery: https://hunchful.io/llms.txt
The Hunchful app itself is a private codebase; this repo documents its public MCP surface. Provided by Fynda Invest AB (Sweden).
來源:README.md,提交 3b5a975
工具
0版本歷史
1- v1.0.0最新Sep 16, 2026
評分
已滿足適用檢查項的 75%
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15 分
請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。
- 描述必需15 分
- 圖示10 分
- 文件15 分
- 授權條款10 分
請在儲存庫或註冊中心條目中宣告授權條款。