Mail MCP
Local MCP server for macOS Mail reads plus visible unsent compose, reply, and forward drafts, and constrained single-message moves.
README
Mail MCP
Local MCP server for macOS Mail reads plus visible unsent compose, reply, and forward drafts, and constrained single-message moves.
Built entirely by OpenAI GPT-5.4 via Codex.
Overview
This project exposes Apple Mail accounts, mailboxes, inbox messages, and direct message reads through MCP.
Supported tools:
mail_check_accessmail_list_accountsmail_list_mailboxesmail_list_mailbox_messagesmail_list_inbox_messagesmail_get_messagemail_compose_messagemail_reply_to_messagemail_forward_messagemail_move_message
The primary read workflow is mail_list_inbox_messages, which returns messages for all inboxes and supports headersOnly: true when you only need headers and metadata.
The compose workflow is intentionally narrow: the server can open visible unsent compose, reply, and forward drafts, but it does not expose any send tool. The move workflow is intentionally constrained: it supports moving one message within the same account, but it does not allow moves to Trash or deleted-message mailboxes.
Path Conventions
- Tool inputs use explicit segment arrays:
mailboxPathSegments: ["Inbox"]mailboxPathSegments: ["Archive", "2026"]destinationMailboxPathSegments: ["Receipts"]
- Mailbox results include both
pathandpathSegments. Usepathfor display andpathSegmentswhen calling tools. - Message results include both
mailboxPathandmailboxPathSegments. UsemailboxPathfor display andmailboxPathSegmentswhen calling tools. - Message results use
id. Tools that operate on a specific message require that same value under the input namemessageId.
Example round trip:
-
Call
mail_list_mailboxesand pick a mailbox result such as:{ "accountId": "account-1", "name": "Receipts", "path": "Archive / 2026 / Receipts", "pathSegments": ["Archive", "2026", "Receipts"] } -
Call
mail_list_mailbox_messageswith:{ "accountId": "account-1", "mailboxPathSegments": ["Archive", "2026", "Receipts"] } -
Take a returned message object such as:
{ "id": 4812, "accountId": "account-1", "mailboxPath": "Archive / 2026 / Receipts", "mailboxPathSegments": ["Archive", "2026", "Receipts"] } -
Call
mail_get_messagewith:{ "accountId": "account-1", "mailboxPathSegments": ["Archive", "2026", "Receipts"], "messageId": 4812 }
Requirements
- Node.js 20+
- macOS
- Mail.app configured with at least one account
- Automation permission to control Mail
- A signed
MailMCPHelperAppbuild
Setup
-
Install dependencies:
npm install -
Copy
.env.exampleto.envif you want to override defaults:cp .env.example .env -
Build the helper app:
npm run build:helper-appThe build script compiles the app unsigned, then signs the finished bundle with the first local
Apple Developmentidentity. Override that selection withMAIL_MCP_CODESIGN_IDENTITYif needed. -
Run local quality checks:
npm run verify -
Run a real Mail smoke test:
npm run test:live
Optional environment variables:
MAIL_MCP_REQUEST_TIMEOUT_MSMAIL_MCP_HELPER_TIMEOUT_MSMAIL_MCP_HELPER_APP_PATH
Run
Start the server in development:
npm run dev
Build production output:
npm run build
npm start
Permissions
The server launches a signed MailMCPHelperApp and runs Mail automation inside that app process. This keeps the macOS Automation/TCC boundary attached to the helper app instead of Terminal or the client process.
Run mail_check_access once after installation. On first use, macOS may prompt for permission to control Mail.
If the prompt does not appear automatically in your client, launch the helper app directly once:
open MailMCPHelperApp/build/Build/Products/Release/MailMCPHelperApp.app
The server starts even before access is granted. Use mail_check_access to confirm permission state and get an actionable error message if macOS Automation consent is still pending.
To re-run the consent flow on a test machine, reset Mail Apple Events permission for the helper app and then run mail_check_access again:
tccutil reset AppleEvents com.andrewfinke.mailmcphelper
After the reset, confirm that the macOS prompt names MailMCPHelperApp, then confirm System Settings shows MailMCPHelperApp -> Mail.
Verify
npm run verify
Notes
- The filesystem-backed
~/Library/Mailapproach is intentionally not used here because it requires broader privacy access and depends on undocumented Mail storage details. - Message bodies are returned for inbox and mailbox listing tools unless
headersOnly: trueis set. - Compose, reply, and forward tools always open visible drafts. Hidden draft mode is intentionally avoided because Mail does not clean up invisible compose objects reliably through automation.
- Raw headers are optional for list tools and default to
false. - Attachment metadata is included, but not attachment bytes.
- The helper app executes JavaScript for Automation in-process via
OSAKit.
MCP Setup
The server runs over stdio.
Example client configuration for the built server:
{
"mcpServers": {
"mail": {
"command": "node",
"args": ["/absolute/path/to/mail-mcp/dist/index.js"],
"cwd": "/absolute/path/to/mail-mcp",
"env": {
"MAIL_MCP_HELPER_APP_PATH": "/absolute/path/to/mail-mcp/MailMCPHelperApp/build/Build/Products/Release/MailMCPHelperApp.app"
}
}
}
}
For local development without building first:
{
"mcpServers": {
"mail": {
"command": "npx",
"args": ["tsx", "src/index.ts"],
"cwd": "/absolute/path/to/mail-mcp"
}
}
}
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。