Zammad MCP Server: Setup, Usage, and What You Can Do with MCP

📖 10 minute read • Updated May 20, 2026

If you search for Zammad MCP, you want one thing: connect an AI assistant to your Zammad helpdesk without writing a custom REST integration. The Zammad MCP Server (MIT, open source) does exactly that — it exposes tickets, users, organizations, and groups as Model Context Protocol tools for Claude Desktop, Cursor, and other MCP clients.

This guide covers setup end to end, what you can do in daily use, and where to go deeper. Full reference docs live on our site: Zammad MCP Server documentation.

GitHub: github.com/Softoft-Orga/zammad-mcp-server
PyPI: zammad-mcp-server

What is the Zammad MCP Server?

The server sits between your MCP client (Claude, Cursor, etc.) and the Zammad REST API:

flowchart LR
  Client[Claude or Cursor]
  MCP[Zammad MCP Server]
  API[Zammad REST API]
  Client -->|MCP tools| MCP --> API

It ships 30+ typed tools, built-in prompts (ticket summary, customer communication, escalation analysis), resource URIs (zammad://ticket/{id}), and access control so you can deny delete operations for agent workflows.

Unlike Zammad 7 native AI inside the Zammad UI, MCP connects external assistants you already use for coding and automation. Unlike Open Ticket AI Runtime, it does not train custom models — it is a free bridge for immediate experimentation.

Prerequisites

Python 3.11+ (or uvx without a global install)
A running Zammad instance (self-hosted, cloud, or Docker dev stack)
A Zammad personal access token

Step 1 — Create a Zammad API token

Open your profile

Click your user icon (bottom-left) → Profile.

Open Profile from the Zammad avatar menu.

Open Token Access

Select Token Access, then click Create.

Zammad Token Access page with Create button.

Name and permissions

Choose a clear name (e.g. MCP_Cursor).
Set expiration if your policy requires it.
Enable scopes you need. For typical agent work, include at least:
- ticket.agent
- user_preferences

Add admin scopes only if you use admin-level MCP tools.

Add a Personal Access Token — name, expiry, and permissions.

Copy the token once

Zammad shows the secret only once. Copy it, then confirm OK, I’ve copied my token.

Copy your new personal access token before closing the dialog.

Store it as ZAMMAD_HTTP_TOKEN — never commit it to git.

Step 2 — Install the MCP server

Fastest path (recommended):

uvx zammad-mcp-server

Or install permanently:

pip install zammad-mcp-server
# or
uv tool install zammad-mcp-server

Step 3 — Connect Claude or Cursor

Add the server to your MCP config. Example for Cursor and Claude Desktop:

{
  "mcpServers": {
    "zammad": {
      "command": "uvx",
      "args": ["zammad-mcp-server"],
      "env": {
        "ZAMMAD_URL": "https://your-zammad.example.com",
        "ZAMMAD_HTTP_TOKEN": "paste_token_here",
        "MCP_DENIED_TOOLS": "delete_ticket,delete_user,delete_organization"
      }
    }
  }
}

Restart the client, then verify:

Run health_check on Zammad.

Details: Claude & Cursor setup · Configuration

What you can do — usage examples

Once connected, you work in natural language; the assistant calls MCP tools against Zammad. Below are practical patterns teams use on day one.

Ticket operations

You ask…	What happens
”List my open tickets.”	`search_tickets` with filters
”Summarize ticket #12345.”	`get_ticket` + `get_ticket_articles` or `ticket_summary_prompt`
”Find tickets about VPN in Support.”	`search_tickets` with query and group
”Add an internal note to ticket 99.”	`create_article` (you review before sending)
“How many tickets were closed last week?”	`get_ticket_stats`

Users and organizations

You ask…	What happens
”Who is the customer on this ticket?”	`get_user` / ticket payload
”Search users with email @acme.com”	`search_users`
”Show organization Acme Corp.”	`search_organizations` / `get_organization`

Safety and governance

You ask…	What happens
”Which MCP tools are enabled?”	`get_allowed_tools`
”What Zammad version are we on?”	`get_server_info`

Deny delete tools in config so agents cannot remove tickets or users by mistake. See Security.

Built-in MCP prompts

The server includes prompts for structured workflows:

Ticket summary — compress long threads for handoffs
Customer communication — draft replies (human approval still required)
Escalation analysis — overdue / escalation context

Full list: Tools reference

Example session (Cursor)

After setup, a support lead might run:

“Run health_check.” → confirms API access.
“Search open tickets in group Support, newest first.” → triage list.
“Get full thread for ticket 1042 and summarize for escalation.” → summary for a manager.
“Draft a polite customer reply acknowledging the delay — do not post yet.” → draft only; agent posts manually in Zammad.

The MCP server does not replace your review process for customer-facing messages.

Zammad MCP vs other options

Approach	Best for
Zammad 7 native AI	AI inside Zammad UI (summaries, writing assistant)
Zammad MCP Server	Claude, Cursor, custom agents via MCP
Open Ticket AI Runtime	Custom-trained routing/classification on-prem

They complement each other. MCP is the fastest way to try AI on live tickets from the tools you already have open.

Troubleshooting

Problem	Fix
401 / 403	Regenerate token; check `ticket.agent` and role
Connection error	`ZAMMAD_URL` must use `https://` and be reachable from your PC
Tool missing	Check `MCP_DENIED_TOOLS` and run `get_allowed_tools`

More: Quick Start troubleshooting

Next steps

Full Zammad MCP Server docs — configuration, deployment, tools
GitHub repository — issues and contributions
Custom AI on your ticket data? Open Ticket AI for Zammad — on-prem models and routing beyond generic LLMs

Built by Open Ticket AI. The Zammad MCP Server is free and open source (MIT); support and implementation services are available via softoft.de for DACH teams.