send_issue_email MCP

Sends the contact email and problem description left when a user is transferred to a human agent to a fixed service mailbox via SMTP.

The Agent only sees one tool: send_issue_email(email, question).

What the three email addresses are

.env is infrastructure, deployed once. email / question differ per conversation and are passed by the Agent from conversation variables (e.g. $邮箱, $问题).

Related MCP server: MCP Sendmail Server

Environment requirements

• Python 3.10+

• Access to the SMTP server in use (direct connection to Gmail from this machine requires outbound 465/587)

Installation

cd send_issue_email_mcp
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env

Fill in .env as described below, do not commit .env to the repository.

Configuring .env

Required: recipient and SMTP

TARGET_EMAIL=issues@example.com

SMTP_HOST=smtp.gmail.com
SMTP_PORT=465
SMTP_USERNAME=your_account@gmail.com
SMTP_PASSWORD=xxxx xxxx xxxx xxxx
SMTP_USE_SSL=true

SENDER_EMAIL=your_account@gmail.com
SENDER_NAME=问题收集机器人

Two equivalent Gmail combinations:

Workspace / school Google accounts may have SMTP disabled by the admin; in that case it won't work.

MCP transport (remote client)

智齿 sends POST / to the root path of the URL you fill in, so the remote defaults are as follows:

MCP_TRANSPORT=streamable-http
MCP_HOST=0.0.0.0
MCP_PORT=8765
MCP_PATH=/

After changing .env, if you have already exported variables with the same names in the terminal, load_dotenv() will not override them. Open a new terminal, or unset MCP_TRANSPORT TARGET_EMAIL before starting.

Verify sending first (without starting MCP)

source .venv/bin/activate
python test_email.py

On success it prints: 成功：问题已发送至服务邮箱 ...

The first argument in the script is the simulated user contact email, not the sender. It's recommended to change it to your own real email for testing, to avoid example.com being blocked into spam by the recipient.

SMTP success only means the post office accepted the mail. If the target is an enterprise Feishu or similar mailbox, also check the sent folder "Sent", for bounce-backs, and the recipient's spam/quarantine.

Start MCP (智齿 / remote)

source .venv/bin/activate
python server.py

The log should show something like:

transport 'streamable-http' (stateless) on http://0.0.0.0:8765/

Do not run two instances at once, otherwise you'll get address already in use. When restarting, stop the old process first, then start.

URL to fill in 智齿

Select Streamable HTTP / HTTP as the transport type (not SSE).

The URL should not include /sse or /mcp; fill in the root address that can reach this machine:

http://<这台机器的可达地址>:<MCP_PORT>/

For example, on LAN: http://192.168.x.x:8765/

If 智齿 is a cloud SaaS, it cannot reach an intranet IP. You need a public IP, domain, or tunnel; the URL is still the root path, without a suffix.

Save and reconnect. The service log should show POST / HTTP/1.1" 200, not 404.

Local self-check

curl -sS -X POST http://127.0.0.1:8765/ \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"check","version":"0.0.1"}}}'

A normal response returns JSON with "jsonrpc":"2.0" and "serverInfo":{"name":"issue-collector"...}.

Start MCP (local stdio)

After setting MCP_TRANSPORT=stdio in .env, local clients like Cursor launch it with a command, for example:

{
  "mcpServers": {
    "issue-collector": {
      "command": "/绝对路径/send_issue_email_mcp/.venv/bin/python",
      "args": ["/绝对路径/send_issue_email_mcp/server.py"]
    }
  }
}

In stdio mode, do not write normal logs to stdout.

Tool conventions

send_issue_email(email: str, question: str) -> str

• email: user contact email

• question: problem/request

• Returns a string starting with 成功： or 错误： for the Agent to branch on

Email subject: [问题收集] 来自用户 {email} ({时间})
The body contains the user email, submission time, and problem description; Reply-To is the user email.

After send_message succeeds, the MCP result is returned immediately; SMTP QUIT/connection close happens in a background thread to avoid SSL close dragging past 智齿's timeout.

Logs

Execution logs are written to log/ under the project, one file per day; multiple starts on the same day append:

log/2026-08-19.log

Includes HTTP request start/end, tool calls, SMTP connect/login/send timing, and background close results. Does not include the SMTP password.

Troubleshooting

load_dotenv() runs once when server.py starts. If you only change .env without restarting the process, the running service won't read the new values.