Multi Mail
co.lumman由 lumman.co 发布官方发布者v1.2.1更新于 Nov 3, 2026
Gmail, Outlook, Hotmail, Microsoft 365, iCloud, Fastmail, other IMAP mailboxes: search, read, send.
概览
Multi Mail 的托管 MCP 服务(mail.lumman.co),由 lumman.co 发布;无需登录;Streamable HTTP。
暂无 AI 概览:以下信息来自注册中心条目和服务本身。
- 发布者
- lumman.co
- 端点
- mail.lumman.co
安装
在 SourceWeft 中
- 打开 控制台中的 Multi Mail,将其添加到工作区。
- 为需要使用其工具的对话启用该服务。
Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。
交给智能体安装
把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。
阅读 https://sourceweft.com/mcp/co.lumman%2Fmail/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。其他 MCP 客户端
把它添加到你客户端的 mcpServers 配置中。
{
"mcpServers": {
"multi-mail": {
"type": "http",
"url": "https://mail.lumman.co/mcp"
}
}
}README
Multi Mail
Gmail, Outlook, Hotmail, Microsoft 365, iCloud, Fastmail, other IMAP mailboxes: search, read, send.
Multi Mail is a remote MCP server run by Lumman Ltd. It puts every mailbox you connect - Gmail, Outlook.com, Hotmail, Microsoft 365, iCloud, Fastmail and other IMAP mailboxes - behind one connector, so an AI client can search all of them in one call and read, draft, send and organise mail in each. This repository holds what clients install: the manifests for Claude Code, Cursor, Gemini CLI, Antigravity and Kiro, four skills and their evals. The server itself is not in it.
Before you start
You need a Lumman account with at least one mailbox connected at https://lumman.co/mcp/account/multi-mail. Sign in there with Google, Microsoft, a passkey or an emailed code. Outlook, Hotmail and Microsoft 365 mailboxes connect with Microsoft sign-in, iCloud and Fastmail with an app password, other IMAP mailboxes with their address and a password, and Gmail with Google sign-in while Google's preview has room; once it is full, a Gmail or Google Workspace mailbox connects with an app password. Each mailbox's permissions - reading, drafts, sending, organising and settings - are set on the same page.
Add it to your client
Claude
Open Multi Mail in Claude's connectors directory, https://claude.ai/directory/connectors/multi-mail, and press Connect. The listing connects at https://lumman.co/mcp/mail.
Claude Code
The plugin adds the connector at https://lumman.co/mcp/mail and the four skills below. The first time a skill reaches the connector, Claude Code asks you to sign in; if it does not, run /mcp and choose the multi-mail server.
Cursor
Add the connector to ~/.cursor/mcp.json:
VS Code
Run MCP: Add Server, choose HTTP, paste https://mail.lumman.co/mcp and name it multi-mail. Or add it to your user mcp.json:
Gemini CLI
The extension adds the connector at https://mail.lumman.co/mcp and the four skills below. Its first call opens Lumman's sign-in in your browser; /mcp auth multi-mail opens it again.
Antigravity
The plugin adds the connector at https://mail.lumman.co/mcp and the four skills below, for the Antigravity CLI, app and IDE alike. To sign in, run /mcp in the CLI, open multi-mail_multi-mail and choose Authenticate, or in the app or IDE press Authenticate next to it under Agent settings, Customizations; approve on the Lumman page that opens, then paste the code Antigravity's page shows back into Antigravity.
Kiro
In the Powers panel press Add Custom Power, choose Import power from GitHub, paste https://github.com/lumman-lab/multi-mail/tree/main/kiro and press Install. The power adds the connector at https://mail.lumman.co/mcp.
ChatGPT
Works fully on ChatGPT Business, Enterprise and Edu.
- For a mailbox you connect with a password, add it first in your account, under Add a mailbox.
- Open chatgpt.com/plugins, press +, then Add custom MCP server.
- Name it Multi Mail, paste
https://mail.lumman.co/mcp, choose OAuth, tick the box and press Create as a plugin. - Install it. On the Lumman page, sign in, add a mailbox if none is connected, and approve.
- In a chat, type @ and choose Multi Mail.
Any other client
Add a remote MCP server at https://mail.lumman.co/mcp: Streamable HTTP, OAuth 2.1 with dynamic client registration. No key or client id is needed.
Signing in
The connector answers a client's first call with where to sign in. The client registers itself with Lumman's authorisation server and opens a Lumman page that names it: sign in, add a mailbox if none is connected, and approve. Every client you approve is listed under AI apps at https://lumman.co/mcp/account/multi-mail, where it can be removed.
Skills
Claude Code, Cursor, Gemini CLI and Antigravity load four skills with the connector, and pick one by themselves when a request matches it. In Claude Code each is also a command, such as /multi-mail:triage.
triage- what is waiting across every mailbox: what needs a reply, what needs a decision, what can wait. Changes nothing.attachments- what a PDF (scanned pages included), Word, Excel or PowerPoint attachment says, read as text.drafting- a new message, reply or forward written as a draft and shown to you, then sent only after you say yes to that draft.labels- duplicate labels (Outlook categories), and in Gmail unused ones, found and a merge proposed, with nothing changed until you agree.
Examples:
- "What in my Outlook and Gmail still needs a reply?"
- "What total does the invoice Dana sent to my Hotmail come to?"
- "Reply to Sam in Outlook that Friday works, and show me before it goes."
Ready requests
The connector offers these requests ready-made; a client shows them as prompts or commands.
needs_reply- What needs a reply. Messages across all your mailboxes still waiting for your answer.waiting_on- Who is waiting on whom. Who waits on your reply, and whose reply you wait on, across all your mailboxes.tidy_newsletters- Tidy newsletters. Find newsletters and promotions across all your mailboxes and archive them.morning_brief- Morning brief. A short summary of every inbox, made to run as a Claude scheduled task.clean_up_inbox- Clean up my inbox. See who fills an inbox, then clear only the groups you choose.keep_tidy- Keep my inbox tidy. Keep the newest few from the senders you choose and archive the rest, set up as a scheduled task.manage_rules- Manage my rules. See the rules that file your incoming mail, and add, change or remove only the ones you choose.
Tools
Where your mail goes
Claude and Claude Code send every call to https://lumman.co/mcp/mail, every other client to https://mail.lumman.co/mcp - two addresses of one server, run by Lumman. Signing in runs through the authorisation server those addresses name. Lumman keeps no copy of your mail: each message and attachment is fetched from its provider when a client asks, its words are screened for prompt attacks by Google Cloud's Model Armor - an attachment asked for as the original file is not - as are the words of every draft and send before they reach the provider, and the mail is passed on only to the client you approved - an attachment's original also to whoever holds the 15-minute link that client is given. What the connector passes on, and to whom, is in the privacy policy: https://lumman.co/privacy.
The drafting skill asks the client to show a draft and wait for your yes. What the connector itself enforces is each mailbox's permission to send, set at https://lumman.co/mcp/account/multi-mail: with sending off, nothing is sent, whatever a client asks.
Limits
- Gmail sign-in is in Google's preview, which admits a fixed number of accounts; once its places are spent, a Gmail or Google Workspace mailbox connects with an app password.
- A clear-out, and any change to a draft, in a mailbox connected with a password needs a mail server that names where it moves each message, as iCloud, Fastmail and Gmail do, and clears a Gmail connected that way out of its inbox alone.
- Microsoft lets an Outlook category take a new colour, never a new name.
- Mailboxes connected with a password - iCloud, Fastmail, Gmail and other IMAP mailboxes - take every tool except filters and the vacation responder. iCloud and Fastmail labels are their folders, as your own mail app shows them, and the labels skill leaves them alone.
- Yahoo, AOL, Yandex and Mail.ru mailboxes are not connected.
- Each account has a daily allowance of tool calls; past it, a call is refused with the time the next one is available.
- A message whose words screen as a prompt attack is withheld from the answer that would carry it, and the rest is still given; anything else that screens so is refused, and a call whose screening cannot finish is closed.
Evals
claude plugin eval . runs four cases, one per skill, and thirteen choose-* cases, against invented Hotmail, Gmail, iCloud and IMAP mailboxes answered from fixed mock files under evals/. Each choose-* case is a request in a person's own words - a mailbox by its provider, every account at once, sending, filing, a rule, an auto-reply, clearing out an inbox - that the connector's tools should take, and three are requests no mail tool should take. They read the tools' own descriptions, saved in evals/mocks/multi-mail/_tools.json; the mocks carry no server instructions. Run on 2 November 2026 with Claude Code 2.1.308 and model claude-sonnet-5-5, three runs per case: all thirteen passed 3 of 3.
To see what the skills add, each of the four skill cases ran in two setups with the connector's tools in place: once with the plugin as shipped, once with its skills/ folder removed (claude plugin eval <copy> --ablation none). Run on 18 October 2026 with Claude Code 2.1.292, model claude-sonnet-5-5 and the CLI's default Haiku judge, three runs per case each way:
- triage - lists the two threads waiting for a reply and nothing else: 3 of 3 with the skills and 3 of 3 without. The skill adds nothing measurable here.
- attachments - gives the total and due date read off the invoice PDF: 3 of 3 with and without. The skill adds nothing measurable here.
- drafting - saves the reply as a draft, shows it and asks before sending: 3 of 3 with, 0 of 3 without; without the skills Claude sent the reply at once in every run.
- labels - proposes the merge and asks before changing anything: 3 of 3 with, 0 of 3 without; without them Claude re-labelled messages before asking in every run.
Overall score 1.00 with the skills and 0.74 without. Three runs per case is a small sample, and the mocks answer the same whatever Claude changes.
Security
Report a vulnerability privately through GitHub, as SECURITY.md describes.
Support
Write to the address under Contact Us in Lumman's privacy policy: https://lumman.co/privacy.
Licence
The files in this repository are under the MIT licence in LICENSE. The Multi Mail service they connect to is provided under Lumman's terms: https://lumman.co/terms.
来源:README.md,提交 cf035a1
工具
0版本历史
5- v1.2.1最新Nov 3, 2026
- v1.1.2Oct 18, 2026
- v1.1.1Oct 16, 2026
- v1.0.2Oct 9, 2026
- v1.0.1Sep 27, 2026
评分
已满足适用检查项的 50%
该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。
- 近 30 天内可连通必需25 分
请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。
- 已声明认证方式必需10 分
请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。
- 已列出工具15 分
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分