How to
How to connect Claude to Zendesk
Claude can summarize a Zendesk queue from a read-leaning connection, but the replies and triage still happen in another tab. A custom connector puts Claude on the actual ticket, with public replies posting only after you approve.
Connect Claude to Zendesk through a custom MCP connector that authenticates as your own agent account over OAuth. Claude then searches and reads tickets freely, drafts replies on the actual ticket, sets status and priority, adds internal notes, and assigns agents, with posting a public reply gated behind a confirmation you approve.
Key takeaways
- A custom connector lets Claude draft replies on the real ticket, not just summarize it in the chat window.
- Triage actions like status, priority, tags, and assignment run inline, while posting a public reply waits for your approval.
- Claude can add internal notes visible only to agents, keeping context on the ticket where the team works.
- The connector inherits your agent account's permissions and never sees tickets or groups you cannot.
Can Claude work Zendesk tickets out of the box?
Partly. The Zendesk connections available to Claude today tend to be read-leaning or cover a thin slice, so Claude can often search and summarize tickets, but posting public replies, changing status or priority, adding internal notes, and assigning agents are limited or absent. Those write actions come from a custom connector.
Reading a queue is the easy half of support. Claude can tell you what came in overnight and which tickets look urgent. But support work is writing: the reply, the status change, the internal note, the reassignment. A read-leaning connection means Claude describes the queue while you still work it by hand.
A custom connector adds the write actions with scoped, human-in-the-loop controls, so outbound moves like posting a public reply always wait for you.
The wider pattern holds beyond this one tool: in our June 2026 review of 264 business tools, 77 percent had no Claude connector at all, and none shipped full read-write out of the box.
How do I connect Claude to Zendesk with write access?
You connect Claude to Zendesk by having a custom MCP connector built against the Zendesk API, authenticated as your own agent account over OAuth with the narrowest scopes the work needs. Once added to Claude, you triage and reply in plain language and approve anything customer-facing.
The setup runs like this:
- 1Tell us which support work Claude should handle: drafting replies, triage, internal notes, assignment.
- 2We build a small MCP server with read tools like
search_ticketsseparated from write tools likeupdate_ticketandadd_internal_note. - 3You authenticate the connector as your own Zendesk agent account over OAuth, so it inherits exactly your permissions.
- 4Claude starts working tickets from your conversations, with
post_replyand other outbound actions gated behind your confirmation.
From then on, a ticket gets handled in one place. Claude reads the conversation, drafts the response, sets the fields, and stages the reply. You review and approve, and it posts.
What can Claude do in Zendesk with full read-write access?
With full read-write access, Claude can draft a public reply on a ticket and post it after you approve, add internal notes visible only to agents, set status, priority, and type, tag and assign tickets to the right agent or group, search open tickets to propose batched responses, and create follow-up tickets.
The write tools map to the queue work itself:
- Draft a public reply on the actual ticket, with the conversation history intact, and post it once you approve.
add_internal_noteleaves agent-only context on the ticket, so handoffs carry the reasoning.update_ticketsets status, priority, and type as you triage, and applies tags.assign_ticketroutes work to the right agent or group.create_ticketopens a follow-up from a resolved conversation so nothing gets dropped.
Batch work is where the leverage compounds: Claude can search open tickets matching a query and propose responses across the set, which you approve one at a time or as a group.
Will Claude reply to customers without my approval?
No. Posting a public reply is a separate, gated write action that only happens after you approve it. Claude can prepare the response, set fields, and stage changes, but outbound actions wait for your confirmation. You decide which updates are automatic and which require a review step.
The important boundary in support is the customer-facing one. Internal moves like notes, tags, and status changes can run freely because they are visible only to your team and easy to reverse. Anything a customer sees stays behind the gate.
Public replies are never automatic
Claude can fully draft the response on the ticket, but the reply only goes out to the customer after you confirm it. There is no background agent answering tickets on its own.
Is it safe to give Claude access to my Zendesk account?
Yes, when the connector is scoped properly. It authenticates through OAuth as your own agent account and inherits your existing Zendesk permissions, so it cannot see tickets, fields, or groups you are not already permitted to work with. Scopes are kept as narrow as the task allows, and outbound actions require your confirmation.
Customer data never flows anywhere your own login could not already take it. The connector is your account plus an approval gate, not a new privileged integration with broad access.