The ThriveDesk MCP connector lets an AI assistant work inside your helpdesk: search conversations, draft and send replies, tag and assign tickets, look up contacts, and read your reports. You decide exactly which of those it may do, and you can take the access back at any time.

MCP (Model Context Protocol) is an open standard for connecting AI assistants to the tools you already use. Connecting ThriveDesk this way means your assistant acts on real tickets instead of guessing from what you paste into a chat window.

Once connected, you can ask your assistant things like:

  • What is still open in the Billing inbox from this week?

  • Summarise ticket #2841 and draft a reply, but do not send it.

  • Tag every unassigned refund request with needs-finance and assign them to me.

  • Look up this customer and tell me what they have contacted us about before.

  • How many conversations did each agent resolve last month?

Before you begin

You need three things:

  • A ThriveDesk account that belongs to at least one workspace. Any role can connect. Your role decides what you are able to grant.

  • An AI client that supports remote MCP servers. This article covers Claude (web, desktop, and Claude Code), ChatGPT, and the popular code editors. The supported clients are listed under Add ThriveDesk to your assistant below.

  • A browser signed into ThriveDesk, or your password to hand. You approve the connection on a ThriveDesk page, not inside the assistant.

There is nothing to install, no API key to generate, and no developer setup. The connection is made with your normal ThriveDesk sign-in.

Set up the connection

Copy the ThriveDesk server address

Your assistant needs one address. It is the same for every ThriveDesk customer:

https://mcp.thrivedesk.com

That is the whole address. No path, no key, nothing to fill in.

Add ThriveDesk to your assistant

Claude on the web or desktop

  1. Open Settings, then Connectors.

  2. Choose Add custom connector.

  3. Name it ThriveDesk and paste https://mcp.thrivedesk.com as the URL.

  4. Choose Add, then Connect. A ThriveDesk window opens.

Claude Code

Run this once in your terminal:

claude mcp add --transport http thrivedesk https://mcp.thrivedesk.com

Then type /mcp inside Claude Code, pick thrivedesk, and choose Authenticate. Your browser opens on the ThriveDesk approval page.

ChatGPT, VS Code, Cursor and other supported clients

Add a remote (HTTP) MCP server pointing at https://mcp.thrivedesk.com. Sign-in is handled automatically, so the client sends you to ThriveDesk to approve it. The menu wording differs between clients and versions, but the address never changes.

Supported clients today are Claude (web, desktop, mobile, and Claude Code), ChatGPT, VS Code, and Cursor. If your client is not listed, contact ThriveDesk support and we will tell you whether it is supported yet.

Approve the connection

ThriveDesk shows an approval page with your name and the name of the assistant asking for access. Three things happen here, in order.

Choose the workspace

If you belong to more than one ThriveDesk workspace, pick the one this connection is for. The connection is tied to that single workspace. To use a second workspace, add a second connector and approve it for that one.

Choose how much access to grant

Pick a preset, then adjust it if you want:

Preset

What it grants

Good for

Read-only

Everything you can view. Changes nothing.

Reporting, triage, asking questions

Standard

What your role can do by default.

Everyday work. Start here.

Full access

Everything you personally can do in this workspace.

Power users who want no friction

Below the presets, every permission is listed by area (Conversations, Contacts, Inboxes, Reports, Live Chat and so on). Turn a whole area on or off, or open it and pick individual permissions. Grant the least you need. You can always reconnect with more later.

Authorize

Choose Authorize. You are returned to your assistant and the connection is live. Choosing Deny cancels it and nothing is shared.

Check the name. The approval page shows which application is asking. If you did not just start this setup, or the name is unfamiliar, choose Deny.

Check that it works

Ask your assistant something harmless that needs real data:

List my ThriveDesk inboxes.

You should get your actual inbox names back. If you do, you are done. If the assistant says it has no ThriveDesk tools, see Troubleshooting below.

What it can do

The connector exposes 28 actions. Your assistant picks the right one for what you asked. Anything your grant does not cover simply is not available to it.

Area

Actions

Conversations

Search and list, open a full thread, reply to the customer, add an internal note, save or delete a draft, schedule a reply for later, set status, snooze, assign, tag, set custom fields, move to another inbox, forward to an external address, create a new conversation

Contacts

Search contacts, open a contact with their recent conversations, create or update a contact

Inboxes

List the inboxes you can access

Live chat

List chats in progress, open a transcript, reply to a visitor, accept a chat, close a chat

Knowledge base

Search your help centre articles and answers

Saved replies

List the saved replies available to you

Notifications

List your notifications, mark them read

Reports

Per-agent conversation and resolution counts, agent and team summaries over a date range

Attachments

Get a temporary download link for a conversation attachment

What reaches the customer. Replying to a conversation, replying in live chat, and forwarding a message all send immediately. Internal notes, drafts, and creating a conversation never reach the customer, because creating only stages the conversation for you to review and send. A scheduled reply is queued and goes out at the time you set.

How permissions work

Three rules, in order of strength. Each one can only narrow access, never widen it.

  1. Your own permissions come first. The approval page only ever offers what you personally hold in that workspace. If your role cannot delete conversations, no preset on that page can grant it, and no assistant connected as you can do it.

  2. Your grant narrows it further. Whatever you left switched off is refused, even though you could do it yourself in the web app.

  3. Inbox membership still applies. The connector reaches the same inboxes you do. An inbox you are not a member of stays invisible to it.

Nothing here changes your account. Granting a permission to a connector does not give it to you, and revoking a connector does not take anything away from you.

Review and revoke access

Everything you have connected is listed in one place inside ThriveDesk:

  1. Open Apps.

  2. Choose the Authorized OAuth Apps tab.

Each card shows the application name and the date you connected. Choose Revoke to cut it off. Access stops immediately and the assistant loses every ThriveDesk action until someone authorizes it again.

Worth a habit. Review this list now and then, and revoke anything you do not recognise or no longer use.

Limits and safeguards

Safeguard

What it means for you

Rate limit

Up to 60 actions a minute. A runaway assistant is throttled, not left to flood your inbox.

Session length

Your assistant refreshes its access quietly in the background. After 90 days without use you approve it again.

Repeat protection

Sending actions accept a request key. When an assistant attaches one and then retries after a timeout, ThriveDesk replays the original result instead of sending a second time.

One workspace per connection

A connection can never reach across into another workspace's data.

Full audit trail

Replies, notes, assignments and status changes appear in the conversation thread under your name, exactly as if you had done them in the app.

A word on judgement. An assistant with send access can email your customers. Until you trust how it behaves, grant Read-only, or ask it to draft rather than send. Drafts wait in the conversation for you to approve.

Troubleshooting

What you see

What to do

The assistant has no ThriveDesk tools at all

The connection was added but never approved. Reopen the connector and choose Connect or Authenticate to finish sign-in.

Your token was not granted the capability required for this tool

You left that permission switched off when approving. Revoke the connection under Apps, then connect again and include it.

Your token is not authorized for this resource

Your own role does not allow the action, or the ticket is in an inbox you are not a member of. Ask an administrator for access.

Conversation not found

Ticket numbers like #2841 are not the same as the internal id. Ask the assistant to search for the ticket first, then act on the result.

This request is not scoped to an organization

The connection lost its workspace. Revoke it and connect again, choosing the workspace on the approval page.

The approval page will not load, or says authorization failed

Approval requests expire after ten minutes. Start the connection again from your assistant.

It worked before and now asks you to sign in again

Normal after a long gap, or after someone revoked the connection. Approve it again.

Still stuck? Contact ThriveDesk support with the name of the assistant and the exact message you saw.

Questions

Does this give an AI company access to all our customer data?

No. The connector responds to specific requests your assistant makes on your behalf, within the permissions you granted, for the one workspace you chose. It does not hand over a copy of your helpdesk, and nothing is transferred in bulk or in the background.

Will my teammates see what the assistant does?

Yes. Replies, notes, assignments and status changes land in the conversation thread under your name, the same as work you do in the app.

Can I connect more than one assistant?

Yes. Each one is approved separately, with its own permissions, and each appears as its own card under Apps so you can revoke them independently.

I have two workspaces. Can one connection cover both?

No, and that is deliberate. Add a second connector and approve it for the other workspace.

Can an administrator connect this for the whole team?

No. Each person connects their own assistant, because every action runs as that person and respects their inbox access. There is no shared or workspace-wide connection.

What happens when I leave the workspace?

The connection stops being able to do anything. It carries your access, so when your membership ends, every action it tries is refused.