Zendesk MCP Server

A Model Context Protocol (MCP) server that puts Zendesk inside your AI assistant, for both sides of the conversation.

For agents: it finds answers in the Help Center; drafts, updates and translates articles while keeping the languages in sync; and handles Support tickets end to end, comments, triage and image attachments included.

For your customers: the Help Center's "Submit a request" form, in plain language: pick the request, answer its questions, follow the replies, close it. See End-user mode.

For your Zendesk admins: teach the assistant how your team works, with house tone, lexicon, procedures and ready-made workflows, from a dedicated back-office, with no redeploy. See Context and customizations.

It all happens in plain language, without switching apps.

It does roughly what the Zendesk agent for Microsoft 365 Copilot does, minus the tie to one vendor: it drops into any MCP client (Claude Desktop, Claude Code, Cursor, VS Code, and the rest). And it always acts with each user's own Zendesk permissions, never a shared admin key.

What your assistant can do

Ask in natural language; the assistant works out the context and the intent, then calls the right tools on your behalf.

• Find answers in the Help Center. "How do I request a software license?" or "what's the time-off policy?" surfaces the right article, matched by meaning rather than by keyword.
• Create, view and update tickets without leaving the conversation: open a ticket, check its status, add a public reply or an internal note, change the priority or the assignee, mark it solved.
• Summarize a ticket for a report or a quick decision. The assistant pulls the details and the full comment thread and gives you the gist in a sentence.
• Read the screenshots and photos attached to a ticket. Error dialogs, UI captures and product photos are handed to your assistant's own model as images, so it can describe them or act on what they show.
• Search and triage your queue in plain language: "show me my open tickets about billing from this week."
• Draft and maintain knowledge-base articles. You can write a new one, or revise a large one a single section at a time, so the whole HTML body never has to round-trip through the model.
• Follow your team's playbook. Instructions, prompts and skills written by your admins reach every user's assistant, whatever MCP client they use.
• Submit and follow a request as a customer, not an agent. Choose between the kinds of request the vendor offers, answer only the questions that form asks, and then track it: what was replied, what you replied, and whether it's done.

Why this server

Most Zendesk integrations run on a shared admin API key, which hands every user full access to every ticket, and bolt on a fixed set of tools. This one is built differently.

• Per-user authentication, OAuth only. Both transports use OAuth 2.1 PKCE: each user signs in with their own Zendesk credentials, so the assistant sees and touches exactly what that person is allowed to, the same scoping you get by signing into Zendesk directly. Static API tokens are deliberately not supported (why).
• Two audiences, one server. Because auth is per-user rather than a shared admin key, the same install serves your agents and your customers — the end-user surface is a namespace you switch on, speaking the API path Zendesk reserves for requesters. Most Zendesk MCP servers are agent-only by construction.
• Section-based article editing. For large Help Center articles, read and rewrite one section at a time (parsed by h1/h2/h3 headings) instead of shuffling the full HTML body through the assistant. On a targeted edit that cuts tokens by a factor of 10 to 100.
• Native multimodal attachments. Ticket images come back as native MCP image content, so the client's own model (Claude, GPT, Gemini, whichever) sees the pixels directly. No server-side vision model, no extra API key, and nothing tying you to one provider. Non-image attachments come back as text references, and both image caps are configurable.
• A tool surface you can cap. Expose every operation as its own tool, group them into namespace proxies, or collapse everything into a single unified tool. You can also filter by namespace or down to read-only operations, so each context loads only the surface it needs (see Tool surface).
• Your practices ship with the server. Most teams hand-distribute a skill or a rules file per AI client. Here admins author instructions, prompts and skills once, inside the server, behind a back-office reserved to Zendesk admins, and every MCP client receives them. Skills fall back to prompts on clients that don't support them yet.
• Two deployment shapes, same auth story, on a lean stack. Run it on your laptop as a stdio MCP server, or deploy it as a private remote MCP server reached over HTTP, each client carrying its own user's token; underneath, the official MCP TypeScript SDK plus zod, speaking to the Zendesk Support and Help Center (Guide) APIs.

Look elsewhere when:

• You need Zendesk products outside Support and Guide (Talk, Explore analytics, Sell). Those endpoints aren't covered.
• You need a single shared service account, or static API-token auth. This server supports neither, by design (see below).

What this server does not do

There is no API-token authentication. The server speaks OAuth 2.1 PKCE and nothing else: no ZENDESK_EMAIL + ZENDESK_API_TOKEN (Basic auth) mode, in any transport. That is deliberate, for two reasons.

1. API tokens are insufficiently secure. A Zendesk API token is a long-lived, static, shared secret that carries the full rights of the user who issued it. There is no per-user scoping, no short expiry, and no per-user consent or revocation. OAuth 2.1 PKCE issues per-user, revocable tokens instead.
2. API tokens don't scale. A single static credential can't attribute actions to individual users, and it can't be revoked granularly. It also makes a multi-user remote deployment unsafe: over HTTP it would expose the issuing user's rights to every caller.

If you specifically need an API-token or service-account mode (headless CI with a shared account, say), use one of the other Zendesk MCP servers that support it. A few are listed under Inspiration & related projects.

Quick start: local (stdio)

The default shape: one developer, one Zendesk account, OAuth 2.1 PKCE in the browser. You need Node.js >= 22 and a Zendesk instance (Support or Suite).

Install

# Run without installing
npx -y @fruggr/zendesk-mcp-server <your-subdomain>

# Or install globally
npm install -g @fruggr/zendesk-mcp-server
zendesk-mcp-server <your-subdomain>

Signing in needs a Zendesk OAuth client, so register one first (next section).

Zendesk OAuth setup

1. Go to Admin Center → Apps and integrations → APIs → OAuth Clients
2. Create a public client:
   • Identifier: <your-subdomain>_zendesk (or set ZENDESK_OAUTH_CLIENT_ID)
   • Redirect URL: http://localhost:27439/callback (change the port to match OAUTH_CALLBACK_PORT / --callback-port if you override it; Zendesk accepts several redirect URLs, one per line)

If the client restricts its allowed scopes, it needs read — plus write unless the server runs with --read-only, which asks Zendesk for the read scope alone.

On the first tool call the server starts the sign-in flow: it opens a browser window and returns the authorize URL in a tool message. The call does not block waiting for sign-in, so authenticate in the browser and then retry the request. The token is persisted to an owner-only file and reused across restarts, so you don't authenticate again every time your MCP client respawns the server (path and overrides: OAUTH_TOKEN_FILE). Several instances can run side by side — a read-write one and a --read-only one, say — without trading credentials.

MCP client wiring

{
  "mcpServers": {
    "zendesk": {
      "command": "npx",
      "args": ["-y", "@fruggr/zendesk-mcp-server", "<your-subdomain>", "--mode", "single"]
    }
  }
}

claude mcp add zendesk -- npx -y @fruggr/zendesk-mcp-server <your-subdomain> --mode single

{
  "servers": {
    "zendesk": {
      "command": "npx",
      "args": ["-y", "@fruggr/zendesk-mcp-server", "<your-subdomain>", "--mode", "single"]
    }
  }
}

Something not working? See Troubleshooting.

Quick start: remote (HTTP)

Experimental. The HTTP transport ships, but it has not yet been exercised end-to-end against a real Zendesk tenant from every MCP client. Local stdio is the supported path.

You can also deploy a private remote MCP server for one Zendesk account. It is its own OAuth authorization server: web clients such as claude.ai and ChatGPT add it by URL and register themselves, each user signs in to Zendesk through it, and no shared admin key or per-client Zendesk setup is involved. Each release also ships a hardened, distroless container image for it (ghcr.io/fruggr/zendesk-mcp-server), signed and attested. The full guide covers the image, the one redirect URL to add to your Zendesk OAuth client, the master secret and grant store, --public-url behind a reverse proxy, trusted clients, MCP client wiring, CORS and what stays the operator's job: docs/http-deployment.md.

Tool surface

Tools are grouped into namespaces: Tickets, Help Center, Users & Organizations, Search, Connector preferences — this server's own per-user settings, such as SLA alert thresholds, never written to Zendesk — and Requests — the end-user surface, which is opt-in and covered under End-user mode. The server registers them in one of three modes, so you can trade granularity against context budget:

• all: every operation as its own tool, for clients with good tool selection;
• namespace (default): one proxy tool per namespace, a balanced middle ground;
• single: a single zendesk tool, for models with limited tool slots (the connector's own preferences keep their connector_preferences proxy beside it).

Proxies take { "operation": "<tool_name>", "params": { … } } and validate params through the original schema. --namespace, --tool and --read-only filter tools before the proxies are built, so each proxy describes only the operations that survive.

There's one way to pick the inventory (--namespace / --tool), --mode packages it, and --read-only narrows it. When the combination isn't obvious, don't guess — --print-tools has the server answer what it would expose, with no credentials needed.

Every tool with its description and its read/write mode: docs/mcp-tools-reference.md. The flags and worked examples: docs/configuration.md.

End-user mode

The audience for everything above is a Zendesk agent. This section is about the other one: your customers.

On the web, a customer opens a ticket through the Help Center's "Submit a request" form — they pick a kind of request, fill in the fields that kind asks for, attach a file, and later come back to read the replies. End-user mode is that same journey, in their assistant.

Who it's for

A vendor pointing their customers at an MCP server, so support happens where those customers already work. They install it themselves, sign in with their own Help Center account, and never see anything that isn't theirs.

Turning it on

The end-user tools live in the requests namespace, and it is opt-in: an agent install shouldn't inherit tools built for someone else, and one of them (marking a request solved) doesn't work under an agent token at all — Zendesk accepts it and silently does nothing. help_center is worth serving alongside it, since a customer who can search the knowledge base often doesn't need to open a ticket in the first place.

The flags, and how to have the server print what a combination exposes: docs/configuration.md.

The journey it supports

• See what kinds of request are available. The forms the vendor offers, by their customer-facing names.
• Learn what one of them asks. The questions on that form, which are required, and for dropdowns the exact choices — so the assistant can gather them in conversation instead of guessing at a payload.
• Submit it, with attachments.
• Follow it. List their requests, read one with its whole conversation (each reply attributed, and support agents marked as such), reply back, and mark it solved when it is.

What it asks of the Zendesk account

Nothing the agent side doesn't already need, plus one thing: at least one ticket form marked visible to end users, because that is what a customer picks between. Sign-in is interactive, through a browser, by design — there is no scripted or headless path to an end-user token, which is the same protection that stops anyone else signing in as your customer.

The prerequisites in full: docs/configuration.md. The step-by-step walkthrough, written for someone who doesn't work in a terminal: docs/end-user-onboarding.md.

What a customer can't do — and shouldn't

A customer can do less than an agent. That's the point, not a gap:

• Only their own tickets. Enforced by Zendesk, not by us.
• No internal notes, in either direction. Agent-only notes never appear in what this surface returns — Zendesk filters them out of the requester's view of a ticket entirely.
• No priority or type. Zendesk drops both when a customer sets them; triage stays with the agents.
• Closing a ticket only once an agent has picked it up. Until then Zendesk won't let the requester solve it, so the tool says so rather than pretending.
• No search across the ticket base, no user lookups, no views, no macros.

Context and customizations

The assistant gets two kinds of context besides tools:

• Built in: the structure of your Help Center (locales, the category → section tree with IDs, segments, permission groups), so it uses real IDs instead of guessing, fetched with the caller's own token. Details: docs/help-center-context.md.
• Yours: instructions, prompts and skills your Zendesk admins write from /mcp/admin, a second connector on the same server. A guided setup prompt walks them through it and reminds them what is already built in. Nothing is stored in Zendesk, and no workflow is imposed: each team writes its own. Details: docs/customizations.md.

Configuration

The complete reference for the CLI flags (--mode, --namespace, --read-only, --transport, --public-url, and so on) and the environment variables (ZENDESK_SUBDOMAIN, OAUTH_TOKEN_FILE, PUBLIC_URL, the attachment-vision caps) lives in docs/configuration.md. Every variable has its own anchor, so you can deep-link a specific setting.

Troubleshooting

Browser not opening during OAuth login, the callback port already in use, having to re-authenticate every time, and Permission denied on the Guide-admin endpoints are covered in docs/troubleshooting.md. Restart with LOG_LEVEL=debug for the full OAuth flow trace.

Development

Setting up the repo, the toolchain, dev mode and how to test a PR branch are covered in CONTRIBUTING.md. Architecture and code-style conventions live in AGENTS.md.

FAQ

Which Zendesk products are supported? Zendesk Support (tickets, users, organizations) and the Help Center / Guide (articles, sections, categories, translations, labels, content tags, segments, attachments). Talk, Explore and Sell are out of scope.

Do I need a Zendesk admin API key? No, and the server doesn't support one. Each user authenticates with their own credentials and the server acts with exactly their permissions (why).

Is it safe to run via npx? Releases are published from CI via npm Trusted Publishing (OIDC), so each version carries a build provenance attestation you can verify on its npm page. No secrets are ever logged by the server.

Contributing

Pull requests are welcome, AI-assisted ones included, as long as the human author has read and validated every line. The guide, the author checklist and the review workflow are in CONTRIBUTING.md.

Versions follow SemVer and are released automatically from Conventional Commits; the history is in CHANGELOG.md.

Inspiration & related projects

This project was built with reference to:

• The official Zendesk API documentation
• mattcoatsworth/zendesk-mcp-server
• koundinya/zd-mcp-server

License

MIT

Built and maintained by Digital4better for the Fruggr project.