fagents-mcp
MCP server that enables Claude Code agents to send and read emails via SMTP/IMAP with per-agent credential isolation and audit logging.
README
fagents-mcp
MCP (Model Context Protocol) server exposing email tools to Claude Code agents. Handles SMTP sending and IMAP reading, with per-agent credential isolation and an audit-logged secure email gate.
Stack: TypeScript, Node.js, Express, @modelcontextprotocol/sdk. Stateless MCP over HTTP.
Tools
| Tool | Description |
|---|---|
send_email |
Send email via SMTP. Supports plain text, HTML, CC/BCC, and file attachments. |
list_emails |
List messages in a mailbox folder. Returns newest first with envelope metadata. |
read_email |
Read a full message by UID — body text/HTML and attachment list. |
search_emails |
Search by from/to/subject/date range/unseen/body text. |
list_mailboxes |
List available IMAP folders. |
download_attachment |
Download an attachment by part number. Returns base64-encoded content. |
gate_email |
Secure read: fetches email via IMAP, logs full content to #email-log channel (human-visible audit trail), returns metadata to caller. Set read_body=true to also receive body — content is always logged first. Use this for all inbound email access. |
Architecture
Claude agent (MCP client)
│ POST /mcp (x-api-key: <agent-token>)
▼
fagents-mcp (Express + MCP SDK)
├─ auth.ts — resolve agent from API key
├─ config.ts — load per-agent env from agents.json, AsyncLocalStorage context
├─ email.ts — SMTP send via nodemailer
├─ imap.ts — IMAP read/search/download via node-imap
└─ server.ts — MCP tool registration, gate_email audit logging
Stateless: Each MCP request creates a fresh server+transport instance. No session state between calls.
Per-agent credentials: Each agent sends its unique API key. The server resolves the agent ID, then loads that agent's credentials (SMTP host/pass, IMAP host/pass, etc.) from agents.json via AsyncLocalStorage. Agents never see each other's credentials.
Credential Storage
Credentials live in agents.json in the working directory (typically the installation directory). This file is managed by the human operator — agents must not read or write it directly.
{
"agents": {
"ftf": {
"apiKey": "<hashed-or-raw-key>",
"SMTP_HOST": "smtp.example.com",
"SMTP_USER": "ftf@example.com",
"SMTP_PASS": "...",
"IMAP_HOST": "imap.example.com",
"IMAP_USER": "ftf@example.com",
"IMAP_PASS": "...",
"COMMS_TOKEN": "...",
"COMMS_URL": "http://127.0.0.1:9754"
}
},
"shared": {
"SMTP_PORT": "587",
"IMAP_PORT": "993"
}
}
AgentConfig accepts any [envKey: string]: string pairs alongside apiKey — the schema is open for extension (Telegram tokens, etc.).
Environment Variables
Without agents.json, the server falls back to environment variables:
| Variable | Default | Description |
|---|---|---|
MCP_PORT |
3000 | Listen port |
MCP_HOST |
127.0.0.1 | Bind address |
MCP_API_KEY |
— | Single-agent API key (if not using agents.json) |
SMTP_HOST |
— | SMTP server hostname |
SMTP_PORT |
587 | SMTP port |
SMTP_USER |
— | SMTP username |
SMTP_PASS |
— | SMTP password |
SMTP_FROM |
SMTP_USER | From address |
IMAP_HOST |
— | IMAP server hostname |
IMAP_PORT |
993 | IMAP port |
IMAP_USER |
— | IMAP username |
IMAP_PASS |
— | IMAP password |
IMAP_TLS |
true | Use TLS for IMAP |
COMMS_URL |
— | fagents-comms URL (for gate_email audit log) |
COMMS_TOKEN |
— | Comms token (for gate_email audit log) |
Setup
npm install
npm run build # compiles TypeScript → dist/
# Single-agent mode (env vars)
MCP_API_KEY=secret SMTP_HOST=smtp.example.com ... node dist/server.js
# Multi-agent mode (agents.json in working dir)
node dist/server.js
Development (no build step):
npm run dev # runs src/server.ts directly via tsx
Claude Code Integration
Add to .claude/settings.json (or via MCP config):
{
"mcpServers": {
"fagents-mcp": {
"type": "http",
"url": "http://127.0.0.1:3000/mcp",
"headers": {
"x-api-key": "<your-agent-api-key>"
}
}
}
}
MCP tools are discovered at session start. After changing tools or restarting the server, start a new Claude session to pick up changes.
Security Notes
- gate_email strips
<untrusted>tag variants from email body before wrapping content — prevents boundary escape attacks from malicious emails. - Subject and messageId are excluded from
gate_emailresponse (attacker-controlled fields that could carry injection payloads). They are logged to#email-logfor human review. agents.jsonshould bechmod 600and never committed to version control.- Server binds
127.0.0.1by default. Use SSH tunnels or a reverse proxy for remote access.
Origin
Part of the fagents project — autonomous Claude Code agent infrastructure. Built by freeturtle agents and Juho Muhonen.
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。