The initdesk MCP server lets any MCP-compatible AI client work directly with your helpdesk: reading tickets, drafting replies, managing customers, and maintaining your knowledge base — all in natural language, without leaving the assistant.

What is MCP?

The Model Context Protocol (MCP) is an open standard for connecting AI assistants to external tools and data. Instead of copying ticket text into a chat window, your assistant talks to initdesk directly through a set of well-defined tools.

Server endpoint

https://mcp.initdesk.com/mcp

The server uses a remote HTTP transport and authenticates via OAuth. When you connect for the first time, you'll be redirected to initdesk to sign in and authorize the client. No API keys to copy or rotate.

The assistant acts on your behalf

This is the most important thing to understand about the connection.

When you authorize the MCP server, you are not creating a separate service account or a machine identity with its own permissions. The server acts as you — the user who completed the OAuth sign-in. Every request the assistant makes is executed under your initdesk identity, with exactly the access level and permissions your account already has.

Concretely:

• Same visibility. The assistant can read the inboxes, tickets, customers, and articles you can read — and nothing else. Inboxes you don't have access to are invisible to it.
• Same write permissions. If your role can't publish articles, delete tickets, or change portal settings, neither can the assistant. Restricted actions fail the same way they would in the web app.
• Same audit trail. Actions taken through MCP are attributed to your user. A reply sent by the assistant appears as sent by you, and internal notes carry your name. There is no "AI" actor in the history.
• Same organization scope. Access is limited to the organizations your account belongs to. If you're a member of more than one, name the organization in your prompt so the assistant works against the right one.
• Revocable at any time. Disconnecting the connector ends the assistant's access immediately. If your initdesk account is deactivated or your role changes, the assistant's effective permissions change with it — automatically, with no separate step.

The practical consequence: the MCP connection never widens your access, but it does mean anything you can do, the assistant can do while connected. Grant it the same trust you would grant a new teammate operating from your own logged-in session.

General setup (any MCP client)

Most clients accept a remote MCP server through a JSON configuration block:

1
2
3
4
5
6
7
8{
  "mcpServers": {
    "initdesk": {
      "type": "url",
      "url": "https://mcp.initdesk.com/mcp"
    }
  }
}

Steps:

1. Open your client's MCP or connectors settings.
2. Add a new remote server with the URL above.
3. Save and restart the client if required.
4. Complete the OAuth sign-in when prompted.
5. Confirm the tools appear — ask the assistant to list your inboxes as a smoke test.

For clients that use a command-line launcher rather than native remote support, you can bridge with mcp-remote:

1
2
3
4
5
6
7
8{
  "mcpServers": {
    "initdesk": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.initdesk.com/mcp"]
    }
  }
}

Setup in Claude (web, desktop, and mobile)

These steps apply to the Claude app, not Claude Code.

1. Open Settings → Connectors.
2. Click Add custom connector.
3. Enter a name — initdesk — and the server URL https://mcp.initdesk.com/mcp.
4. Click Add, then Connect next to the new entry.
5. Sign in to initdesk in the OAuth window and approve access. Whichever initdesk account you sign in with here is the identity every subsequent action runs under.
6. The connector now shows as connected. In a new chat, open the tools menu to confirm the initdesk tools are enabled.

A few notes specific to Claude:

• Custom connectors are available on paid plans. On Team and Enterprise, an administrator may need to enable custom connectors for the workspace before individual members can add them.
• The connector is tied to your personal Claude account and your personal initdesk login. It is not shared with teammates — each person connects with their own credentials and gets their own permission scope.
• You can toggle the connector per conversation from the tools menu, which is useful when you want a chat that doesn't touch production data.
• Claude asks for confirmation before write operations. Approve individually while you're getting comfortable, then switch to allowing the tool for the rest of the chat.
• Mention initdesk explicitly in your prompt ("check initdesk for open tickets tagged billing") so Claude reaches for the right tool.

What the server exposes

• Tickets — list, search, retrieve, create, reply, add internal notes, assign and unassign.
• Customers — list, retrieve, create, and update requesters.
• Knowledge base — list, search, create, and update articles and collections.
• Configuration — inboxes, tags, ticket statuses, saved filters, and portal settings.

Every one of these is filtered by your permissions. Search supports full-text and faceted queries, so you can ask for things like "unresolved tickets from the last 14 days mentioning refunds" and get a real result set rather than a guess.

Five things worth trying

1. Morning triage in one prompt. Ask for every unassigned ticket opened since yesterday, grouped by topic, with a one-line summary and a suggested priority for each. You get a triage plan before you've opened the inbox.

2. Turn resolved tickets into help center articles. Point the assistant at a cluster of tickets that all asked the same question, and have it draft a knowledge base article covering the answer. Create it as a draft, review, then publish. Each article written this way removes a recurring ticket permanently.

3. Find gaps in your knowledge base. Ask the assistant to compare the most common ticket subjects from the last quarter against your published article titles and report which topics customers keep asking about but have no article. It's a prioritized content backlog derived from real demand.

4. Draft consistent replies at volume. Give the assistant a batch of similar tickets and ask it to prepare replies in your tone, each pulling the relevant article link. Review the drafts, adjust, and send. Useful after an incident when twenty people report the same thing.

5. Answer questions about your own support data. Ask how response times shifted month over month, which tags grew fastest, or which customers opened the most tickets this quarter. Because the assistant queries live data, you get the answer conversationally instead of exporting a CSV.

Troubleshooting

• Tools don't appear — restart the client and verify the connector shows as connected in settings.
• Authorization expired — disconnect and reconnect the connector to run OAuth again.
• Empty results — check that you're pointing at the right organization; if you belong to more than one, name it in your prompt.
• Write operations fail — confirm your initdesk role allows the action. The MCP server enforces the same permissions as the web app, so if you can't do it in the browser, the assistant can't do it either.

A note on safety

Because the assistant operates with your identity and your permissions, treat the connection as you would handing someone your logged-in session. Review write operations before approving them, especially anything that replies to customers or publishes articles. Content inside tickets comes from external senders, so keep a human in the loop before acting on instructions found in ticket text.