Zendesk MCP: How to Connect Zendesk to AI in 2026
Yes — Zendesk now runs a first-party MCP server, and it lives on your own subdomain rather than a shared vendor endpoint. It answers at https://<your-subdomain>.zendesk.com/api/mcp, authenticates over OAuth, and its published metadata advertises both read and write scopes, so an AI client can do more than look.
Worth separating two Zendesk things that share a name, because they point in opposite directions. The MCP client — announced at Zendesk’s Relate conference in May 2026 and opened for early access in June — lets Zendesk’s own AI agents reach out to other systems like Asana, Sentry, and Stripe. The MCP server is the direction most people mean by “connect Zendesk to AI”: it lets Claude, ChatGPT, or Cursor reach into your Zendesk. Both now exist, and confusing them will send you down the wrong setup path.
Alongside the first-party server, several community-built MCP servers still wrap the public Zendesk API — they predate Zendesk’s own and remain an option if you want to self-host or shape the tool surface yourself. For most people the first-party server is now the shorter path. And worth knowing before you set any of it up — MCP, official or not, hands Zendesk to an AI inside a conversation you start. It’s a doorway, not a worker. Here’s what’s actually available, how to connect it, where it stops, and what to use when you want Zendesk work that runs without you.
Looking beyond Zendesk? MCP servers by category covers the rest.
What an MCP connection to Zendesk does
Model Context Protocol (MCP) is the open standard that lets an AI client — Claude, ChatGPT, Cursor, and others — talk to an outside app through a shared interface. Zendesk now serves that interface itself, at https://<your-subdomain>.zendesk.com/api/mcp, behind OAuth and advertising read and write scopes. Community projects like reminia/zendesk-mcp-server and mattcoatsworth/zendesk-mcp-server on GitHub, plus a mcp-zendesk package on PyPI, wrap the Zendesk Support API the same way and predate the first-party endpoint. With either connected, an AI client can:
- Look up tickets — “show me the open tickets from Acme Corp” answered from your live instance.
- Search and filter — pull tickets by tag, status, priority, or requester without opening Zendesk.
- Draft responses — read a ticket’s history and comments, then write a reply grounded in the actual conversation.
- Create and update — file a new ticket, add a comment, change a status from inside the chat.
That’s genuinely useful for ad-hoc support work — ask a question, get an answer pulled from real tickets, make an edit on the spot. The difference now is that you can get it from Zendesk’s own endpoint instead of somebody else’s code.
How to connect Zendesk to AI right now
The short path is Zendesk’s own server. The community route is still there if you want to control the tool surface yourself.
- Point your AI client at your own subdomain. In Claude, ChatGPT, or Cursor, add a custom MCP connector with the URL
https://<your-subdomain>.zendesk.com/api/mcp— your real Zendesk subdomain, not a shared vendor host. - Authorize over OAuth. The endpoint supports dynamic client registration, so the client can register itself and walk you through the consent screen without you hand-creating an API token. Its published metadata lists
readandwrite, so read the consent screen before you approve. - Ask a ticket question — “show me open tickets from Acme Corp” — to confirm tools are visible and the scopes you granted work.
- Or take the community route if you’d rather self-host.
mattcoatsworth/zendesk-mcp-servercovers Support, Talk, Chat, and Guide;reminia/zendesk-mcp-serverfocuses on tickets and comments;mcp-zendeskon PyPI is a lighter Python option. These need a Zendesk API token (Admin Center → Apps and integrations → APIs → Zendesk API) plus your subdomain and agent email, and you host the process. None are Zendesk-maintained, so check recent commits before trusting one with write access.
Separately, if what you actually want is Zendesk’s AI agents reaching out to Asana, Sentry, or Stripe, that’s the MCP client — a different product, opened for early access in June 2026 through Zendesk’s EAP waitlist.
Or skip hosting anything and go through the plain Zendesk API directly with a script or a tool like Carly, covered below.
Where the Zendesk MCP stops
A first-party server closes the trust question, not the shape of the protocol. MCP works the same way everywhere:
- It only works inside a chat you start. Close the window and nothing happens. Nobody is watching your ticket queue; the AI waits for you to ask.
- No triggers. A new ticket coming in, an SLA about to breach, a CSAT score dropping — none of these can start anything through MCP, official server or not.
There is no such thing as an MCP trigger. The protocol's own working group states plainly that clients find out about server-side changes by polling for them, and a real event mechanism remains unshipped.
- It’s one app at a time. A Zendesk MCP server knows Zendesk. Getting an escalated ticket into Slack, a CRM, and a spreadsheet means wiring up a separate MCP server for each and hoping your client can juggle them in one turn.
- The scopes are yours to reason about. Zendesk’s endpoint advertises write access, so what you approve on the OAuth screen is the blast radius against your support data. With a community server, add your own API token and hosting to that list.
So even with Zendesk’s own server, an MCP setup is a way to ask about tickets and make one-off edits. It’s not a way to make Zendesk run — to have work happen on a schedule or in reaction to a ticket event, across the other tools support touches.
Running Zendesk work that doesn’t need a chat open
That “run on its own, across apps” gap is exactly where Carly fits. Carly connects to Zendesk natively — no MCP server to connect, no API token to babysit — and to the ~260 other apps it supports, plus anything with a public API through your own key. The difference from MCP is the important part: Carly’s workflows are triggered and scheduled, so Zendesk work happens whether or not anyone has a chat window open.
A few things MCP can’t do but a Carly workflow can:
- When a ticket is tagged “urgent” → post it to the on-call Slack channel, create a linked Jira issue, and text the on-call engineer — automatically, the moment it’s tagged.
- Every morning → summarize tickets open more than 48 hours with no reply and send the list to the support lead.
- When a CSAT survey comes back low → pull the ticket transcript, draft a follow-up for the agent to review, and log the case in a tracking sheet.
The non-AI steps — the moving, matching, and routing between apps — are free, the Zapier-style backbone of the workflow. The AI steps (drafting, summarizing, deciding) start at $35/month. You describe the outcome in plain language and Carly wires up the Zendesk connection and everything downstream.
If you just want to interrogate your ticket queue from a chat, Zendesk’s own MCP server (or a community one) can do that. If you want Zendesk to actually do things — on a trigger, on a schedule, across every tool a ticket flows through — that’s the job MCP wasn’t built for, and it’s the one Carly was.
FAQ
Does Zendesk have an official MCP server?
Yes. It runs per-tenant at https://<your-subdomain>.zendesk.com/api/mcp rather than on a shared vendor host, answers over OAuth with dynamic client registration, and publishes read and write scopes. Don’t confuse it with the MCP client Zendesk announced at Relate in May 2026 and opened for early access in June — that one points the other way, letting Zendesk’s own AI agents reach systems like Asana, Sentry, and Stripe.
Is there any way to connect Zendesk to AI via MCP today?
Yes — start with Zendesk’s own endpoint on your subdomain. Community-built servers like reminia/zendesk-mcp-server, mattcoatsworth/zendesk-mcp-server, or the mcp-zendesk PyPI package still work and are worth it if you want to self-host or trim the tool surface. None of those are maintained by Zendesk, so treat them like any third-party code touching support data — review before granting write access.
Can a Zendesk MCP server trigger automations? No. MCP is request/response inside an AI chat — it has no triggers and nothing runs when the conversation is closed. For event- or schedule-driven Zendesk work across apps, you need a workflow tool like Carly rather than an MCP server.
Can I connect Zendesk to AI without coding or hosting a server? Yes. You don’t have to touch MCP, official or community, at all. Carly connects to Zendesk for you and lets you build the automation in plain language — describe what you want to happen and it wires up Zendesk and the other apps involved, with no server to host and no code to write.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
