Messages Bridge

io.github.IsaiahDupree由 GitHub 帳號 IsaiahDupree 發布v1.3.0更新於 Oct 24, 2026

MCP connector for iMessage & Contacts via a local Mac agent + Vercel relay

已驗證Streamable HTTP可網頁執行效率與工作流溝通與協作

概覽

AI 產生的概覽

這個連接器讓 ChatGPT 透過執行在你自己的 Mac 上的小型代理，搜尋和傳送 iMessage 對話，並讀取或新增聯絡人。

可以做什麼
搜尋訊息，列出或分頁瀏覽最近的會話與對話，並可檢查本機存取狀態與涵蓋範圍而不回傳訊息內容。向單一收件人傳送一則 iMessage，送出前需你確認。依姓名、電話或電子郵件搜尋、讀取和新增聯絡人。讀取在你的 Mac 本機執行，請求經託管中繼短暫排隊轉送。
適用情境
適合讓助理查找過去的 iMessage 對話或聯絡人資訊，或起草並傳送單則簡訊，同時讓訊息資料留在自己的 Mac 上。需要一台保持喚醒並執行代理的 Mac。
執行需求
遠端 MCP 端點，需 OAuth 登入、託管服務帳號和配對碼。一台裝有 Node 18+ 與 Apple 命令列工具的 Mac，執行 apple-messages-agent 命令列工具，並授予完整磁碟存取權限以及 Messages 和聯絡人的自動化權限。Mac 須保持喚醒且代理持續執行。
安裝前請注意
代理會讀取本機完整訊息資料庫，並可傳送訊息和新增聯絡人；兩個寫入工具標記為破壞性操作，需你確認，且每次只傳給一個收件人。託管中繼會在短效期佇列中暫時保存請求與結果內容，訊息內容在傳輸中離開你的 Mac。代理權杖保存在本機檔案中；自架需要 Supabase 與 JWT 密鑰。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Messages Bridge，將其新增到工作區。
  2. 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。
  3. 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.github.IsaiahDupree%2Fmessages-bridge/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。 客戶端會在首次連接時完成 OAuth 登入。

{
  "mcpServers": {
    "messages-bridge": {
      "type": "http",
      "url": "https://messagesbridge.vercel.app/mcp"
    }
  }
}

README

💬 MessagesBridge

Use your iMessage & Contacts inside ChatGPT — through your own Mac.

[License: MIT] [npm]

MessagesBridge is a Model Context Protocol (MCP) connector that lets ChatGPT work with your iMessage conversations and Contacts. Reads execute on your own Mac through a small local agent. Requested results pass through the hosted relay and are returned to ChatGPT. The relay temporarily stores request/result payloads in its queue; it does not maintain a permanent copy of your message history.

  ChatGPT  ──OAuth──►  MessagesBridge relay  ──job queue──►  apple-messages-agent  ──►  Messages.app + Contacts (connector)          (Vercel, stateless)      (your Mac, polls & executes)        (on your Mac)

Your Mac is the only place your messages are ever read or contacts written. The relay stores only short-lived job payloads and your account record; it never sees a message unless a job is in flight, and queue payloads have short expiry windows.

For users — 3 steps

1. Create your account at messagesbridge.vercel.app and generate a pairing code.

2. Install the Mac agent (needs Node 18+ and Apple Command Line Tools):

bash
npx apple-messages-agent pair <YOUR-CODE>   # link this Mac to your accountnpx apple-messages-agent install            # keep it running on every restart

The agent needs two macOS permissions, granted once: Full Disk Access so it can read your message history (System Settings → Privacy & Security → Full Disk Access), and Automation the first time it sends a message or reads Contacts — macOS asks "Terminal wants to control Messages" and to access Contacts; click OK.

3. Connect ChatGPT: Settings → Plugins → turn on Developer mode (Security & login) → Create a custom plugin → paste the MCP Server URL https://messagesbridge.vercel.app/mcp → Authentication OAuth → sign in & Allow.

That's it. In a new chat: "Search my messages for the dinner plans" or "Text Sarah that I'm running 10 minutes late." (You confirm every send before it goes out.)

Your Mac must be awake with the agent running. npx apple-messages-agent install makes it start automatically at login and restart if it ever crashes.

Managing the agent

bash
npx apple-messages-agent status      # is it paired & reachable?npx apple-messages-agent logs        # recent activitynpx apple-messages-agent uninstall   # stop auto-start

What ChatGPT can do

ToolAction
search_messagesSearch plain and Apple attributed text; follow next_cursor even on empty result pages
list_recent_threadsList recent conversations, newest first, with a last-message preview
get_threadRead pages of one conversation using an exact chat id; pass next_cursor as cursor until null
messages_statusCheck local access, counts and coverage without returning message content
send_messageSend an iMessage to one recipient (you confirm first — no bulk)
search_contactsFind people by name (returns their phones + emails)
get_contactRead one contact by name or id
create_contactAdd a contact (name, optional phone/email)

Reads come first; the only write tools are send_message and create_contact, and each is confirmed by you in ChatGPT before it runs. Sends always go to one recipient at a time — never in bulk.

Self-hosting

You can run your own relay so nothing depends on the hosted instance.

Prereqs: a Vercel account and a Supabase project (free tiers are fine).

  1. Storage — in your Supabase project's SQL editor, run supabase/schema.sql. It creates isolated actp_messagesbridge_* tables and mb_* storage functions with RLS and service-role-only grants.
  2. Deploy the server/ directory to Vercel.
  3. Set env vars (see server/.env.example):
    • SUPABASE_URL, SUPABASE_SERVICE_ROLE_KEY — your project + service-role key
    • JWT_SECRET — openssl rand -hex 32
  4. Verify: curl https://YOUR-APP.vercel.app/api/health → redisConfigured, redisOk, jwtSecretSet all true.
  5. Point the agent at it: npx apple-messages-agent pair <CODE> --server https://YOUR-APP.vercel.app.

How it works

  • Auth — OAuth 2.1 with PKCE and Dynamic Client Registration (RFC 7591). ChatGPT registers itself, the user signs in on the MessagesBridge consent page, and ChatGPT gets a scoped MCP access token. JWTs are HMAC-signed; three kinds (session, agent, mcp) with distinct privileges. Optional email verification via Resend (RESEND_API_KEY); off unless configured, and enforcement is opt-in (REQUIRE_EMAIL_VERIFICATION).
  • Relay — MCP tool call → job pushed to jobs:<user> (TTL ≤ the caller's wait window, so an abandoned job can't run late) → the Mac agent (holding a hanging long-poll) is handed the job within ~200ms, executes it, pushes the result → the MCP handler returns it. The long-poll means jobs are delivered on arrival rather than on a fixed interval, so relay overhead is ~300ms instead of ~1s. The agent is considered "online" only while it's actively connected.
  • Agent — a dependency-free Node CLI. Read tools query the Messages chat.db directly (read-only SQLite); send and contact tools run via JXA (osascript -l JavaScript) against Messages.app and Contacts; arguments are passed as JSON over argv (never shell-interpolated). A macOS LaunchAgent keeps it alive across logins and crashes.

Repository layout

server/    Vercel functions: OAuth + MCP endpoint + relay        (deploy this)agent/     apple-messages-agent — the npm-published Mac CLI          (users npx this)kit/       Browser automations: register the dev-mode connector AND           fill the OpenAI directory submission (see kit/README.md)supabase/  schema.sql for self-hosting the storagetest/      end-to-end OAuth + MCP integration test

Development

bash
# servercd server && cp .env.example .env.local   # fill in, then: vercel dev# end-to-end test against a live deployment (reads ../.env.local)node test/e2e-oauth.mjs# agent unit testsnode --test agent/test-agent-unit.mjs

Security & privacy

  • Messages are read and sent, and contacts read and added, only on your Mac. The relay temporarily persists requested content in its job/result queues with short TTLs; results are consumed on delivery.
  • The agent token is stored at ~/.messagesbridge-agent.json (mode 600).
  • No secrets are committed; see .gitignore and the .env.example files.
  • Write tools are marked destructive so ChatGPT asks before it sends a message or adds a contact — every send goes to one confirmed recipient at a time, never in bulk.

License

MIT © Isaiah Dupree

Verified read coverage and live tests

Agent 1.1 reads the live SQLite WAL, decodes Apple attributed strings using macOS Foundation, and provides bounded cursor pagination. Cursors follow database insertion order. Attachment-only records remain visible through has_attachments; attachment files are not uploaded or decoded. The Mac database is the coverage boundary: this does not establish that every iPhone/iCloud message has synced.

sh
apple-messages-agent doctornode test/live-messages.mjs# After pairing your own account and starting the agent:MB_EMAIL=your-email MB_PASSWORD=your-password node test/live-relay.mjs

Provide credentials through your secure environment, not a shared transcript or committed file. The live tests only read and print aggregate results. live-messages checks every local conversation page, all messages in the largest thread, decoded search, and cursor validation. live-relay validates actual OAuth, MCP and paired-Mac reads; it fails if the Mac is offline. Reviewer/demo output is never accepted as evidence of a live integration.

Production requires the storage migration and valid Supabase configuration. /api/health returns HTTP 503 when storage is unavailable. If the old relay database is unavailable, accounts, pairing and OAuth sessions from it cannot be recovered automatically; sign in/create an owner account and reconnect the plugin.

來源：README.md，提交 0ecb5ed

工具

0
該伺服器需要登入，因此 SourceWeft 無法在沒有帳號的情況下讀取其工具，也未收錄。已連接帳號所能看到的內容可能因帳號及其權限而異。

版本歷史

1
  1. v1.3.0最新Sep 16, 2026

評分

已滿足適用檢查項的 85%

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