tokenless-zendesk-mcp
You can sign in to Zendesk once in a browser and then use the saved session to read and modify Zendesk tickets via MCP tools.
Sign in with any method (password/SSO/2FA), save cookies/CSRF, and check session validity.
Read views, fetch view tickets, search tickets/users, get ticket details, requester/organization tickets, ticket fields, metrics, macros, user details, and full audit trails.
Make arbitrary read-only GET requests to any Zendesk
/api/v2endpoint.Write to tickets: add public/internal comments, update ticket fields/custom fields, and apply macros (requires CSRF; confirm before sending).
Download ticket attachments using the authenticated session (held back while GDPR sanitization is on).
Get all tool output PII-sanitized by default, with configurable allowlist, threshold, pseudonyms, and kept entity types.
Provides tools for interacting with Zendesk, including reading tickets, views, users, macros, and performing writes like adding comments, updating tickets, and applying macros, using session-based authentication.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@tokenless-zendesk-mcp list my active views in Zendesk"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Tokenless Zendesk MCP Server
An MCP server for Zendesk. You sign in once in a real browser (Playwright) — any method works, including SSO and 2FA — and the resulting session cookies are then used to call the Zendesk REST API directly. No API token, OAuth client, or admin setup required: it acts with exactly your agent account's permissions.
This is the same way Zendesk's own agent web UI talks to its API — a valid session
cookie is all a GET request needs (no CSRF token is required for reads).
What it can do
Reads
Tool | Description |
| Open a visible browser to sign in (password / SSO / 2FA). Saves the session cookies + CSRF token. |
| Check whether a saved session exists and is still valid (calls the API as the current user). |
| List the agent's active views with numeric IDs, titles, and cached ticket counts. |
| Fetch tickets in a view, following pagination ( |
| Search via |
| Subject, status, parties, tags, custom fields (incl. product), full comment thread, and attachments. |
| Download an attachment (by its |
| List the tickets a user has requested (their history) — "have they reported this before?". |
| List tickets belonging to an organization. |
| Field definitions with valid dropdown/tagger option values — look up the |
| SLA / timing metrics for a ticket (reply time, resolution time, reopens, replies). |
| List active macros with ids and titles (for |
| Find users by free-text (name, email, …). |
| Fetch one user by id. |
| Full audit trail (every change/event) for a ticket. |
| Read-only passthrough to any |
Writes (require the CSRF token — see Writes below)
Tool | Description |
| Add a public reply or internal note to a ticket. |
| Set status, priority, type, assignee, group, tags, and custom fields (optionally with a comment). |
| Apply a macro and persist its changes (status/fields/comment) to a ticket. |
Attachments. zendesk_get_ticket returns an attachments array; each item has
an id, file_name, content_url, content_type, and size. Pass an attachment's
content_url to zendesk_download_attachment along with a destination (a file
path, or a directory to save under the original filename). Downloads use the
authenticated session, so private attachments work. While GDPR Compliance
Content Sanitization is enabled,
downloads are held back (see below).
zendesk_request unlocks the rest of the Zendesk REST
API for reads — e.g.
tickets/123/audits, ticket_metrics, organizations, users/123,
satisfaction_ratings. It is GET-only by design (write operations would require a
CSRF token, which is out of scope).
Related MCP server: zendesk-mcp
How auth works
You sign in once in a real browser window. Playwright saves the resulting
cookies to ~/.zendesk-mcp/storageState.json. Every other tool reads the cookies
scoped to your instance and replays them as a Cookie header against
https://{subdomain}.zendesk.com/api/v2/…. When the session expires the API
returns 401/403 and the tool returns a clear "run zendesk_login" message.
The session file holds live auth cookies — it's git-ignored. Treat it like a password.
Writes (CSRF)
GET requests authenticate with the session cookie alone. Write requests
(POST/PUT/DELETE) additionally require Zendesk's CSRF token — the same
one the agent UI uses. It's captured at login (from the agent page's
<meta name="csrf-token">) and saved to ~/.zendesk-mcp/csrf.txt. The token is
stable for the life of the session; if it ever goes stale, the server fetches a
fresh one automatically and retries the write once. The write tools
(zendesk_add_comment, zendesk_update_ticket, zendesk_apply_macro) modify
real tickets — confirm changes with the user before sending.
GDPR Compliance Content Sanitization
On by default. Before any tool result is handed to the LLM, every string in it is scanned for personally identifiable information and each detected value is replaced with a placeholder naming its type:
"Hi, I'm Marie Dubois, email marie@company.fr, tel +33 6 12 34 56 78"
→ "Hi, I'm [PERSON_NAME_1], email [EMAIL_ADDRESS_1], tel [PHONE_NUMBER_1]"Placeholders are numbered per distinct value and stay the same for the life
of the server process. Marie Dubois is [PERSON_NAME_1] in every ticket, search
result and comment, so the LLM can tell people apart and follow them across
tickets without seeing who they are. The value → placeholder mapping lives only
in memory and is never returned by a tool. Numbering is per value, ignoring case
and treating spaces, hyphens, underscores and dots alike. So "Hans-Mueller" and
"hans mueller" match, but "Thomas" alone gets a different number from "Thomas Becker".
Detection runs locally. It uses bardsai/eu-pii-anonimization-multilang (Apache-2.0), a multilingual PII model for the 24 EU languages with 35 entity types. These include names, contact details, addresses, identity documents, bank and card numbers, and the GDPR Art. 9 special categories (health, biometric, ethnic origin, political opinions, …). It runs in-process from the model's ONNX weights via transformers.js. Ticket content is never sent to a third party for this.
Safety nets. Emails, IBANs, payment card numbers (Luhn-checked), IPv4 addresses and international phone numbers are also matched by pattern. Any value detected once in a result is redacted everywhere else in that result too. For example, a requester name recognized in its own field is also removed from comment bodies.
Numeric ids are kept, so the LLM can still chain tools, e.g. from
requester_idtozendesk_get_user. Money amounts are detected but kept by default (seeZENDESK_GDPR_KEEP_ENTITIES).URLs keep working. A redacted Zendesk URL, such as a
next_pagelink with a search query, is mapped back to the real URL when the LLM passes it tozendesk_request.Attachments are held back. Files can't be sanitized yet, so
zendesk_download_attachmentdownloads nothing while sanitization is on. Otherwise the original file would land on disk, where an LLM with file access could read it. Attachment metadata inzendesk_get_ticket(with sanitized filenames) is still returned.Fails closed. If the model can't be loaded, tools return an error instead of unredacted data.
The model (~280 MB for the default q8 weights) downloads on first use into
~/.zendesk-mcp/models. The server starts loading it in the background at
startup. To download it ahead of time instead:
npx -y tokenless-zendesk-mcp download-modelSanitization adds CPU time: roughly 0.1–0.5 s for a typical ticket, and more for
large searches or long threads. Results are cached per string. Detection is
statistical, so it lowers the risk of exposing personal data but can't guarantee
that nothing slips through. Names inside filenames (passport_john_smith.pdf)
are often missed.
Tuning:
Allowlist. The model tags some product and company names as
PROPER_NAMEorORGANIZATION_NAME. List terms that must never be redacted, such as your own products or a shared support mailbox, inZENDESK_GDPR_ALLOWLIST, e.g.ZENDESK_GDPR_ALLOWLIST="SEPTEO Avocat,support@septeo.com". Matching is case-insensitive and whole-word. Only the allowlisted part of a detection is kept, so "Zendesk" on the list still leaves the rest of "Zendesk Suite" redacted.Threshold. A token is redacted when the model's probability that it is PII reaches
ZENDESK_GDPR_THRESHOLD(default0.5). Lower values redact more, mostly by adding false positives: in testing,0.3and0.15started redacting "Zendesk" and "Microsoft Teams" without catching any of the misses. Higher values redact less.Plain placeholders.
ZENDESK_GDPR_PSEUDONYMS=falseswitches back to un-numbered[PERSON_NAME]placeholders.
Set ZENDESK_GDPR_SANITIZATION=false to turn sanitization off.
Add to Claude Code
No clone, no build — just point your MCP config at npx. Add to your .mcp.json
(or Claude Desktop config), replacing youracme with your Zendesk subdomain:
{
"mcpServers": {
"zendesk": {
"command": "npx",
"args": ["-y", "tokenless-zendesk-mcp"],
"env": {
"ZENDESK_SUBDOMAIN": "youracme"
}
}
}
}Or with the CLI: claude mcp add zendesk -e ZENDESK_SUBDOMAIN=youracme -- npx -y tokenless-zendesk-mcp
First login
You sign in once in a real browser window. Run the bundled login subcommand
(it downloads Chromium on first use, then opens the window):
ZENDESK_SUBDOMAIN=youracme npx -y tokenless-zendesk-mcp loginmacOS / Windows / Linux desktop: a Chromium window opens — sign in, done.
WSL2: you need WSLg (Windows 11) or an X server so the window can show. If the window can't open, run the login on the Windows host, or set
DISPLAY.
You can also trigger login from inside Claude with the zendesk_login tool, but
the standalone command is more reliable since not every MCP host surfaces the window.
From source (contributors)
npm install
npm run build
ZENDESK_SUBDOMAIN=youracme npm run login # one-time browser sign-inThen point your MCP config at the build instead of npx:
{
"mcpServers": {
"zendesk": {
"command": "node",
"args": ["/path/to/tokenless-zendesk-mcp/dist/index.js"],
"env": { "ZENDESK_SUBDOMAIN": "youracme" }
}
}
}See CONTRIBUTING.md for the release process and how publishing is configured.
Configuration (env vars)
Var | Default | Purpose |
| — (required) | The |
|
| Where the saved session lives. |
|
| Per-request timeout in ms. |
|
| How long the login window waits for you to finish signing in, in ms. |
|
| GDPR Compliance Content Sanitization — redact PII from all tool output. |
|
| Hugging Face id of the ONNX PII model. |
|
| Weight variant to load: |
|
| Where the model is cached (default: |
|
| Comma-separated entity types to leave unredacted (empty = redact everything). |
| — | Comma-separated terms never redacted, e.g. product names (case-insensitive, whole-word). |
|
| Probability (0–1) a token must have of being PII to be redacted. Lower = redact more. |
|
| Numbered placeholders ( |
Verify
After building and logging in:
ZENDESK_SUBDOMAIN=youracme node scripts/verify.mjsIt calls me, listViews, search, and getTicket against your live instance.
Notes & limitations
Auth is session-cookie based. Cookies expire (and SSO sessions time out), so you'll re-run
zendesk_loginperiodically.It respects whatever permissions your agent account has — nothing more.
The API is paginated; list tools cap results (
limit) and page 100 at a time. Rate-limited (429) responses are retried honoringRetry-After.Writes are supported via the session CSRF token (see Writes). They act with your agent permissions — review changes before sending.
Available Tools
19 toolszendesk_add_commentAdd a comment to a ticketA
Add a comment to a ticket. By default the comment is PUBLIC (visible to the requester); set isPublic=false for an internal note. This modifies the ticket — confirm the wording with the user before sending. Returns the updated ticket.
Inputs: ticketId (required), body (required), isPublic.
zendesk_apply_macroApply a macro to a ticketA
Apply a macro (canned action) to a ticket and persist the resulting changes (status, fields, and/or a comment). Get a macro id from zendesk_list_macros. This modifies the ticket — confirm with the user first. Returns the changes applied and the updated ticket.
Inputs: ticketId (required), macroId (required).
zendesk_download_attachmentDownload a ticket attachmentA
Download an attachment to the local filesystem using the authenticated Zendesk session. Provide the attachment content_url from zendesk_get_ticket and a destination. If destination is a directory, the attachment's own filename is used; otherwise it is the full file path.
Inputs: url (required), destination (required).
zendesk_fetch_view_ticketsFetch tickets in a viewARead-only
Fetch tickets listed in a Zendesk view, following pagination. Returns ticket id, subject, status, priority, requester, assignee, group, and timestamps. Get a view id from zendesk_list_views.
Inputs: viewId (required), limit.
zendesk_get_ticketGet ticket detailsARead-only
Fetch full details for one ticket: subject, description, status, priority, type, tags, requester/assignee/submitter, group, organization, custom fields (with titles, including the product field when present), the full comment thread with authors and timestamps, and a list of attachments. Each attachment includes a content_url — pass it to zendesk_download_attachment to save the file.
Inputs: ticketId (required).
zendesk_get_userGet a userARead-only
Fetch one user by id (name, email, role, organization).
Inputs: userId (required).
zendesk_list_macrosList macrosARead-only
List active macros (canned actions) with their ids and titles. Use a returned id with zendesk_apply_macro.
Inputs: no named parameters.
zendesk_list_viewsList Zendesk viewsARead-only
List the agent's active ticket views with their numeric IDs, titles, and cached ticket counts. Use a returned id with zendesk_fetch_view_tickets.
Inputs: no named parameters.
zendesk_loginLog in to ZendeskA
Open a VISIBLE browser window so you can sign in to Zendesk (password, SSO, and 2FA all work). On success the session cookies are saved and reused by every other tool to call the Zendesk REST API. Run this first, and again whenever a tool reports the session has expired.
Inputs: no named parameters.
zendesk_organization_ticketsList an organization's ticketsARead-only
List tickets belonging to an organization, most recent first. Get an organization id from a ticket (organization field) or a user record.
Inputs: organizationId (required), limit.
zendesk_requestCall a Zendesk API endpoint (read-only)ARead-only
Make a read-only (GET) request to any Zendesk REST API v2 endpoint and return the parsed JSON. Use for anything the dedicated tools don't cover (audits, satisfaction ratings, organizations, users, ticket metrics, etc.). path may be a bare resource like 'tickets/123/audits' or a full '/api/v2/...' path. Pass query parameters via query. See https://developer.zendesk.com/api-reference/ for available endpoints.
Inputs: path (required), query.
zendesk_requester_ticketsList a requester's ticketsARead-only
List tickets requested by a user (their reported tickets), most recent first — useful for seeing whether a customer has raised this before. Get a user id from zendesk_get_ticket (requester) or zendesk_search_users.
Inputs: userId (required), limit.
zendesk_searchSearch ZendeskARead-only
Search Zendesk via the REST API (/api/v2/search). Accepts any Zendesk search query string (e.g. 'type:ticket status:open assignee:me'). Returns matching results, each tagged with its result_type (ticket, user, organization, group) and the raw API record.
Inputs: query (required), limit.
zendesk_search_usersSearch usersARead-only
Find users by a free-text query (name, email, phone, etc.). Returns id, name, email, role, and organization.
Inputs: query (required), limit.
zendesk_session_statusCheck Zendesk session statusARead-only
Report whether a saved Zendesk session exists and verify it is still valid by calling the API as the current user. Use to decide if zendesk_login is needed.
Inputs: no named parameters.
zendesk_ticket_auditsGet ticket audit trailARead-only
Fetch the full audit trail for a ticket: every change and event (field updates, status changes, comments, notifications) with authors and timestamps. More complete than the comment thread alone.
Inputs: ticketId (required), limit.
zendesk_ticket_fieldsList ticket field definitionsARead-only
List ticket field definitions, including the valid dropdown/tagger option values. Use this to find the exact value to pass when setting a custom field via zendesk_update_ticket (e.g. the sentiment or product field).
Inputs: activeOnly.
zendesk_ticket_metricsGet ticket SLA / timing metricsARead-only
Fetch timing and SLA metrics for one ticket: first reply time, full/agent resolution time, reopens, replies, and assignee/requester wait times.
Inputs: ticketId (required).
zendesk_update_ticketUpdate a ticketA
Update fields on a ticket: status, priority, type, assignee, group, tags, and custom fields. For custom field values, look up the exact value with zendesk_ticket_fields. Optionally include a comment. This modifies the ticket — confirm changes with the user first. Returns the updated ticket.
Inputs: ticketId (required), status, priority, type, +6 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 19 tools
Each tool targets a distinct Zendesk action (e.g., add_comment vs update_ticket, list_views vs fetch_view_tickets) and descriptions clarify boundaries. There is no ambiguity between tools.
Nearly all tools follow a verb_noun pattern (zendesk_<verb>_<noun>), with a few exceptions like 'ticket_audits' (noun_noun) but they are still readable and predictable overall.
19 tools cover the core Zendesk workflows—authentication, ticket management, views, macros, search, users, attachments, audits, fields, metrics—without being overwhelming. The count is well-scoped.
The tool set covers most common operations (read, update, comment, search, lists) but notably lacks a create_ticket tool, which is a fundamental gap. The generic request tool is read-only, so creation is not possible.
Maintenance
Related MCP Connectors
Zendesk MCP server for ChatGPT, Claude and any AI agent: list tickets, reply to ticket, update ticket, apply macro, add internal note and search tickets in your own Zendesk account. Cotlet works in the account through your Cloud Browser, a real Chrome where you sign in to Zendesk yourself; Cotlet does not store your password and you need no Zendesk API key. About the official Zendesk API: Zendesk has a full REST API, the Support API (also called the Ticketing API), but no documented Zendesk-hosted MCP server for working tickets (checked 16 Oct 2026). Zendesk's MCP documentation covers the MCP client, where Zendesk calls other servers. Setup and FAQ: https://cotlet.dev/integrations/zendesk-api-mcp-server
Zendesk MCP Pack — tickets, users, organizations via OAuth.
Remote MCP server for managing Muninx tickets, messages, ticket search, and support analytics.
Related MCP Servers
- FlicenseDqualityDmaintenanceA comprehensive MCP server for managing Zendesk Support, Talk, Chat, and Guide products, providing tools for tickets, users, organizations, search, and more.4930-
- AlicenseNot gradedqualityDmaintenanceMCP server for Zendesk support workflows that enables ticket search, lookup, creation, commenting, status updates, and user listing through natural language. It uses OAuth authentication and supports both internal notes and public comments.MIT
- AlicenseNot gradedqualityDmaintenanceMCP server to retrieve information from Zendesk Help Center using anonymous authentication. Provides tools to list categories, sections, articles, and search articles.MIT
- AlicenseBqualityDmaintenanceAn MCP server that provides operational tooling over the Zendesk Support API — ticket triage, SLA monitoring, agent workload analysis, and system health audits.111MIT