eml-mcp
MCP server for managing email archives as .eml files, synced with OneDrive, enabling AI assistants to read/write inbox, outbox, and drafts.
README
eml-mcp
MCP server for managing email archives stored as .eml files. Designed to work alongside a pair of Power Automate flows that automatically export Outlook emails to OneDrive as .eml files, giving AI assistants full read and write access to your inbox, sent items, and drafts through a structured set of tools.
Requirements
- Node.js 20+
- A OneDrive folder (synced locally) —
inbox/,outbox/, anddrafts/sub-directories are created automatically on first run
Installation
npx eml-mcp /path/to/emails --from=you@example.com
Add to your MCP client
<details> <summary>Claude Code</summary>
claude mcp add eml -- npx -y eml-mcp /path/to/emails --from=you@example.com
</details>
<details> <summary>GitHub Copilot</summary>
Add to ~/.copilot/mcp-config.json:
{
"mcpServers": {
"eml": {
"type": "stdio",
"command": "node",
"args": ["/absolute/path/to/eml-mcp/dist/server.js", "/path/to/emails", "--from=you@example.com"]
}
}
}
</details>
<details> <summary>Codex CLI</summary>
Add to ~/.codex/config.toml:
[mcp_servers.eml]
command = "npx"
args = ["-y", "eml-mcp", "/path/to/emails", "--from=you@example.com"]
</details>
Arguments
| Argument | Required | Description |
|---|---|---|
<email-directory> |
Yes | Root directory; inbox/, outbox/, and drafts/ sub-directories must exist inside |
--db-path=<path> |
No | SQLite index path (default: ~/.eml-mcp/index.db) |
--from=<address> |
No | From address for composed drafts (default: draft@eml-mcp) |
Power Automate — Automatic .eml archiving
Each flow watches one Outlook folder and saves every email as an .eml file to the corresponding OneDrive directory. Set up the inbox flow first, then duplicate it for sent items — the only two values that change are noted in the second section.
<details> <summary>Inbox flow — archive received emails</summary>
Trigger
When a new email arrives (V3) — Office 365 Outlook
Set the folder to Inbox. Add subject/sender filters as needed.
Step 1 — Export email (V2)
Export email (V2) — Office 365 Outlook
- Message Id:
Message Id(dynamic value from trigger)
This action returns the raw MIME content of the email (RFC-2822 format, compatible with .eml).
Step 2 — Create file
Create file — OneDrive for Business
- Folder Path:
/Outlook/inbox - File Name:
<Subject>-<Received Time>.eml— compose using dynamic values from the trigger, sanitizing characters not allowed in file names - File Content:
Body(dynamic value from the Export email (V2) step)
The file is created in OneDrive and, on next refresh_index call or server restart, it will be picked up automatically by eml-mcp tagged as inbox.
</details>
<details> <summary>Sent Items flow — archive sent emails</summary>
Duplicate the inbox flow and change only two values:
- Trigger folder:
Sent Items - Folder Path (Create file step):
/Outlook/outbox
eml-mcp will tag these emails as outbox.
</details>
Tools
| Tool | Description |
|---|---|
search_emails |
Full-text search with filters (from, date, attachments, folder…) |
get_email |
Parse and return a single email |
compose_email |
Create a new draft .eml and open it |
update_email |
Modify an existing draft and re-open it |
open_email |
Open an existing .eml in the default mail client (works for inbox, outbox, and drafts) |
delete_email |
Permanently delete an .eml file and remove it from the index |
extract_attachments |
Save one or all attachments from an email to a directory |
open_attachment |
Open an attachment with the system default application |
search_attachments |
Find emails containing attachments by filename, type, or keyword |
refresh_index |
Incrementally sync the index with disk state (add new, remove deleted, update changed) |
Development build
npm install
npm run build
Register the local build in Claude Code:
claude mcp add eml node "/absolute/path/to/eml-mcp/dist/server.js" -- "/path/to/emails" --from=you@example.com
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。