postbote
An MCP server that enables AI assistants to read-only search and retrieve GNOME mail, contacts, and calendar data using IMAP and Evolution Data Server, with no password storage.
README
Postbote
Your GNOME mail, contacts and calendar — on the command line, and as an MCP server so an AI assistant can search your mailbox for you.
Postbote reads the accounts you already configured in GNOME Settings → Online Accounts. There is nothing to log into and no password to store: credentials come from GNOME Online Accounts at runtime and are never written to disk, never logged, and never returned by any command.
It runs on GJS (GNOME's JavaScript runtime) via gjsify — the same stack a GNOME desktop app is built on, which is where this is headed.
Status: early. The CLI and the MCP server work; the desktop app does not exist yet.
What it does
- Mail (IMAP) — search across folders by sender, recipient, subject, date range and full text; read a message; list its parts; save an attachment.
- Contacts and calendar — read the address books and calendars that Evolution Data Server keeps in sync for your online accounts.
- Local index — an optional SQLite full-text index so repeated searches are instant and work offline. Indexing ~1200 messages takes about half a minute; searching them afterwards takes under a second.
Everything is read-only. Messages are fetched with IMAP BODY.PEEK, so
opening a mail through Postbote never marks it as read.
Requirements
- GNOME Online Accounts + Evolution Data Server (Fedora:
gnome-online-accounts,evolution-data-server), andlibgda-sqlitefor the index - A running user session D-Bus — the GOA and EDS daemons are reached over it, so a bare SSH session without one will report the backend as unavailable
- An Email (IMAP/SMTP) account in GNOME Settings. Nextcloud/ownCloud accounts expose files, calendar and contacts but no mail.
- Implicit TLS (port 993). STARTTLS on port 143 is not implemented yet.
Install and run
gjsify install
gjsify workspace postbote-cli build
gjsify run app/dist/postbote.gjs.mjs accounts
Use it
postbote check # which backends are reachable
postbote accounts # which online accounts are available
postbote folders # mailboxes, with their roles
postbote search "energieberater" --since 2025-01-01
postbote search --from berater --all-folders --limit 20
postbote message <uid> --account <id> # one message: body + attachment list
postbote parts <uid> --account <id> # what is attached, and how big
postbote save <uid> --account <id> # write the attachment to disk
postbote sync # build the local index
postbote index status # what it holds, and how fresh
postbote index search "wärmepumpe" # offline, no server contact
postbote contacts --query maier
postbote calendar --from 2026-09-01 --to 2026-09-30
Every command prints JSON — the same shapes the MCP tools return.
search returns headers only, never bodies; reading one message is a separate,
explicit call, and getting an attachment's bytes a third. That is not a policy
you can flip with a flag — the search path contains no code that can fetch a
body.
--since and --before filter the message Date header, not its arrival
time. After a mailbox migration every message's arrival timestamp is the
migration date, which makes an arrival filter useless; --received-since is
there when you genuinely mean arrival.
Search folds diacritics, so marz finds März. (ß is a letter rather than a
diacritic, so grusse does not find Grüße.)
As an MCP server
postbote mcp speaks MCP over stdio. Registered in an MCP client it exposes
mail_search, mail_get_message, mail_list_folders, mail_list_parts,
mail_save_attachment, mail_search_local, mail_sync_status,
contacts_search, calendar_list_events and accounts_list.
The server is read-only by default and fails closed: a tool is registered only if it declares itself read-only, so a future tool that forgets the annotation is silently withheld rather than silently exposed.
See .mcp.json for a working registration.
Your data stays yours
The local index contains message headers and plain-text bodies. It lives at
$XDG_DATA_HOME/postbote/index.db (mode 0600), never inside this repository,
and only postbote sync ever writes to it — a search never does. Attachments
are saved to your download directory. Both locations are overridable via
POSTBOTE_DATA_DIR, POSTBOTE_DB_PATH and POSTBOTE_ATTACHMENTS_DIR.
Nothing is sent anywhere. Postbote talks to your mail server and to nothing else.
Development
See AGENTS.md.
License
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。