Agent Academy

Agent Academy is a free, source-linked library of skills, prompts, MCP servers and bot templates, organised into courses by profession, for people who hand real work to AI agents: 1,467 live resources, each with its original link, license and safety tags.

MCP server

io.github.michaelrice/zendesk-mcp

Zendesk MCP server for Claude Code and other MCP clients

No credentialsDoes not send

https://github.com/michaelrice/zendesk-mcp

Install
uvx zendesk-mcp

Original content

Full originalApache-2.0Source

Republished in full under its license, with attribution. The source is authoritative.

zendesk-mcp

A Model Context Protocol server that exposes Zendesk ticket read and write tools to Claude Code and other MCP clients.

What it does

  • Search, list (paginated), and fetch Zendesk tickets, comments, and attachments
  • Create new tickets and update existing ticket fields (including group, custom status, and tags)
  • Post public replies and internal notes
  • Set ticket status and assign tickets to agents
  • Browse and apply views and macros
  • Look up users, groups, organizations, and custom statuses
  • Read and write time-tracking entries
  • Format a ticket as a Markdown issue draft for handoff to a tracker (GitLab, GitHub, Jira)
  • Two MCP prompts (analyze-ticket, draft-ticket-response) for ticket analysis and response drafting
  • (Optional) Expose Zendesk Help Center articles as an MCP resource
  • (Optional) Read linked GitLab issues / MRs / commits via the Git-Zen Zendesk app

Prerequisites

  • Python 3.10 or newer
  • A Zendesk OAuth client. A Zendesk admin can create one at:

https://<your-subdomain>.zendesk.com/admin/apps-integrations/apis/zendesk-api/oauth_clients

Set the redirect URL to http://localhost:8787/callback and request scopes read write.

Install

Install into a project-local virtualenv. Using a venv keeps zendesk-mcp and its dependencies isolated from your system Python and from other projects, and is the recommended path for everything below.

From a clone of this repository:

python3 -m venv .venv
.venv/bin/pip install --upgrade pip
.venv/bin/pip install -e .

For development (also installs pytest):

.venv/bin/pip install -e ".[dev]"

> Throughout this README, commands use the venv's binaries via .venv/bin/.... You can instead source .venv/bin/activate once per shell and drop the prefix — the result is the same.

OAuth setup

Run the interactive setup using the venv's Python:

.venv/bin/python -m zendesk_mcp setup

You will be prompted for:

  1. Your Zendesk subdomain (e.g. acme for acme.zendesk.com)
  2. The OAuth client ID created by your admin
  3. The OAuth client secret
  4. (Optional) A Git-Zen integration field ID — see [Optional: Git-Zen integration](#optional-git-zen-integration)
  5. (Optional) Whether to enable the Help Center knowledge base resource — see [Optional: Help Center knowledge base](#optional-help-center-knowledge-base)

The setup opens a browser for the OAuth authorization step, then writes a token to ~/.config/zendesk-mcp/config.json (mode 0600).

If you have no browser, the URL is printed to the terminal — open it on any device, click Allow, and paste the resulting redirect URL back into the prompt.

Token expiry and refresh

Zendesk access tokens expire. OAuth clients created on or after 2026-04-30 get a

30-minute default lifetime; older clients issue non-expiring tokens unless an expiry is

requested. Setup requests a 24-hour access token and a 90-day refresh token so the

behaviour is the same either way, and the server renews the access token automatically —

before it expires, and again if Zendesk rejects a token mid-request.

To make that possible, the config file also stores refresh_token, expires_at,

client_id, and client_secret alongside the access token. Keep the file at mode 0600;

it is the same trust level as the access token itself. If your OAuth client returns no

refresh token, setup says so and the token is used as-is.

Re-run .venv/bin/python -m zendesk_mcp setup when:

  • the refresh token expires (90 days with no use), or
  • you revoke the OAuth grant in Zendesk.

In either case the tools return Zendesk authorization failed: ... Re-run: zendesk-mcp setup

rather than failing opaquely.

Register with Claude Code

Register the MCP server using the venv's Python by absolute path. Claude Code launches the server in a fresh shell that does not inherit your activated venv, so the absolute path is required — pointing at a bare python here will fail to import zendesk_mcp.

ZENDESK_MCP_DIR="$(pwd)"   # run this from the repo root, after install
claude mcp add --scope user zendesk -- "$ZENDESK_MCP_DIR/.venv/bin/python" -m zendesk_mcp

Or just inline the absolute path you want:

claude mcp add --scope user zendesk -- /absolute/path/to/zendesk-mcp/.venv/bin/python -m zendesk_mcp

Then add the read tools to permissions.allow in ~/.claude/settings.json to avoid per-call prompts:

{
  "permissions": {
    "allow": [
      "mcp__zendesk__zendesk_get_ticket",
      "mcp__zendesk__zendesk_get_tickets",
      "mcp__zendesk__zendesk_get_comments",
      "mcp__zendesk__zendesk_list_attachments",
      "mcp__zendesk__zendesk_download_attachment",
      "mcp__zendesk__zendesk_search_tickets",
      "mcp__zendesk__zendesk_ticket_to_gitlab_context",
      "mcp__zendesk__zendesk_list_views",
      "mcp__zendesk__zendesk_get_view",
      "mcp__zendesk__zendesk_get_view_tickets",
      "mcp__zendesk__zendesk_list_macros",
      "mcp__zendesk__zendesk_preview_macro",
      "mcp__zendesk__zendesk_search_users",
      "mcp__zendesk__zendesk_get_groups",
      "mcp__zendesk__zendesk_get_group_users",
      "mcp__zendesk__zendesk_get_organization",
      "mcp__zendesk__zendesk_list_custom_statuses"
    ]
  }
}

Write tools (zendesk_post_comment, zendesk_post_internal_note, zendesk_set_ticket_status, zendesk_assign_ticket, zendesk_create_ticket, zendesk_update_ticket, zendesk_log_time, zendesk_add_tag, zendesk_remove_tag, zendesk_apply_macro) are intentionally not in the default allow-list — Claude will prompt you per call.

Restricting tools

Every tool is enabled by default. To narrow what a deployment can do, for example where

ticket text is written by customers and the agent should not be able to reply to them, set

either of these in ~/.config/zendesk-mcp/config.json or in the server's environment. Values

from both are combined.

Config keyEnvironment variableEffect
"read_only": trueZENDESK_MCP_READ_ONLY=trueRegisters no write tools (comments, notes, status, assignment, create/update, tags, macros, time logging)
"disabled_tools": ["zendesk_post_comment"]ZENDESK_MCP_DISABLED_TOOLS=zendesk_post_comment,zendesk_apply_macroDoes not register the named tools

A disabled tool is never registered, so the model cannot see or call it. Names that match no

tool are reported on stderr.

Tools

Tickets

Tools that return ticket records include channel, the channel the ticket arrived on (Zendesk's via.channel: web, email, api, ...), or null when Zendesk does not say. zendesk_get_ticket also returns custom_fields as a list of {id, value} pairs; a field that is not set on the ticket has a null value.

ToolWhat it does
zendesk_search_ticketsSearch tickets by status, priority, type, assignee, requester, tags, or keyword. Each result includes channel
zendesk_get_ticketsList tickets with pagination and sorting (page, per_page, sort_by, sort_order). Each ticket includes channel
zendesk_get_ticketGet one ticket's metadata, including channel and custom_fields
zendesk_create_ticketCreate a new ticket (subject, description, optional priority/type/assignee_id/requester_id/tags/custom_fields)
zendesk_update_ticketUpdate one or more fields on an existing ticket (status, priority, subject, type, assignee_id, requester_id, group_id, custom_status_id, tags, custom_fields, due_at)
zendesk_get_commentsGet the conversation thread on a ticket
zendesk_list_attachmentsList attachments on a ticket
zendesk_download_attachmentDownload an attachment to a local cache directory. The URL must be on your Zendesk host, and dest_dir must be inside the cache or a directory listed in attachment_allowed_dest_dirs in the config
zendesk_ticket_to_gitlab_contextFormat a ticket and its conversation as a Markdown issue draft
zendesk_post_commentPost a public reply on a ticket
zendesk_post_internal_notePost an agent-only internal note on a ticket
zendesk_set_ticket_statusSet ticket status (new, open, pending, hold, solved, closed)
zendesk_assign_ticketAssign a ticket to an agent by email or me

Tags

ToolWhat it does
zendesk_add_tagAdd a tag to a ticket (idempotent)
zendesk_remove_tagRemove a tag from a ticket (idempotent)

Views & Macros

ToolWhat it does
zendesk_list_viewsList all active views
zendesk_get_viewGet a view's filter conditions and execution settings
zendesk_get_view_ticketsFetch tickets currently matching a view. Each ticket includes channel
zendesk_list_macrosList active macros with their actions
zendesk_preview_macroPreview what changes a macro would make
zendesk_apply_macroApply a macro to a ticket (applies field changes and posts any comment)

Users, Groups & Organizations

ToolWhat it does
zendesk_search_usersFind users by name or email
zendesk_get_groupsList all active groups
zendesk_get_group_usersList the members of a group
zendesk_get_organizationFetch an organization including custom fields
zendesk_list_custom_statusesList all custom ticket statuses and their IDs

Time tracking

ToolWhat it does
zendesk_get_time_trackingRead time-tracking entries for a ticket
zendesk_log_timeLog a time entry against a ticket

Git-Zen integration

ToolWhat it does
zendesk_get_git_zen_links(Git-Zen only) Get linked GitLab issues / MRs / commits for a ticket

Prompts

The server exposes two MCP prompts that some clients (e.g. Claude Desktop) surface as slash commands:

PromptArgumentWhat it does
analyze-ticketticket_idAsks the model to fetch the ticket and produce a summary, status/timeline, and key interaction points
draft-ticket-responseticket_idAsks the model to fetch the ticket and draft a customer-facing response (with a confirmation step before posting)

Optional: Git-Zen integration

If your Zendesk instance uses the Git-Zen app, the zendesk_get_git_zen_links tool can read its custom-field payload. Find your instance's Git-Zen custom field ID under Admin → Tickets → Fields (it is a numeric ID), then either set it during .venv/bin/python -m zendesk_mcp setup or edit ~/.config/zendesk-mcp/config.json to add:

{
  "git_zen_field_id": 12345678901234
}

Without this configured, zendesk_get_git_zen_links returns a "not configured" message.

Optional: Help Center knowledge base

If your Zendesk instance has a published Help Center, you can expose its sections and articles as the zendesk://knowledge-base MCP resource. The resource returns a single JSON document covering all sections and articles, cached for one hour.

This is opt-in. Enable it by either answering "y" to the prompt during .venv/bin/python -m zendesk_mcp setup, or by adding the following to ~/.config/zendesk-mcp/config.json:

{
  "knowledge_base_enabled": true
}

When the flag is absent or false, the resource is not registered, keeping the server's resource list empty for instances without a Help Center.

Optional: downloading attachments outside the cache

zendesk_download_attachment writes files to ~/.cache/zendesk-mcp/attachments/<ticket_id>/ by default (configurable via attachment_cache_dir). The dest_dir argument can redirect a download elsewhere, but because attachment contents are controlled by whoever wrote the ticket, the tool only accepts a dest_dir inside the cache or inside a directory you have explicitly allowed. Add the directories you want to permit to ~/.config/zendesk-mcp/config.json:

{
  "attachment_allowed_dest_dirs": ["/home/me/work/support_bundles"]
}

Paths are resolved before the check, so .. segments and symlinks cannot escape an allowed directory. Any dest_dir outside the allowed set is refused before the download starts.

The download URL itself must be an https URL on .zendesk.com for your configured subdomain; the tool refuses anything else so the OAuth token is never sent to another host. Archives are limited to 10,000 entries and 256 MiB unpacked, and tar members that are not regular files or directories are skipped.

Development

python3 -m venv .venv
.venv/bin/pip install -e ".[dev]"
.venv/bin/pytest

Tests run on Python 3.10, 3.11, and 3.12 in CI (see .github/workflows/test.yml).

Releasing

Bump the version in pyproject.toml, mcpb/pyproject.toml (both the version and

the zendesk-mcp== pin), mcpb/manifest.json, and server.json (both fields),

then push a v* tag. That triggers .github/workflows/release.yml, which publishes

to PyPI, packs the MCPB bundle, publishes server.json to the MCP Registry, and

cuts the GitHub release.

Check the versions agree before tagging — the release fails fast otherwise:

python3 .github/scripts/check_versions.py 0.1.5

License

[Apache-2.0](LICENSE)

DescriptionZendesk MCP server for Claude Code and other MCP clients
Version0.1.6
Packagespypi · zendesk-mcp@0.1.6 · stdio
Environment variablesNone stated

Related

TutorialCustomer service

2_1_support_ticket_agent

A structured output agent demonstrating Pydantic schema-based responses for customer support ticket creation.

Sourcegithub.com/Shubhamsaboo/awesome-llm-apps/tree/main/ai_agent_framework_crash_course/openai_sdk_crash_course/2_structured_output_agent/2_1_support_ticket_agent

No credentialsDoes not sendApache-2.0
TutorialCustomer service

3_1_customer_support_ticket_agent

A tutorial demonstrating how to implement a structured customer support ticketing system using Google's ADK (Agent Development Kit) framework. This example shows how to create type-safe, structured support tickets with priority levels, categories, and resolution estimates using Pydantic schemas and

Sourcegithub.com/Shubhamsaboo/awesome-llm-apps/tree/main/ai_agent_framework_crash_course/google_adk_crash_course/3_structured_output_agent/3_1_customer_support_ticket_agent

No credentialsDoes not sendApache-2.0
CaseCustomer service

ai_customer_support_agent

This Streamlit app implements an AI-powered customer support agent for synthetic data generated using GPT-4o. The agent uses OpenAI's GPT-4o model and maintains a memory of past interactions using the Mem0 library with Qdrant as the vector store.

Sourcegithub.com/Shubhamsaboo/awesome-llm-apps/tree/main/advanced_ai_agents/single_agent_apps/ai_customer_support_agent

No credentialsDoes not sendApache-2.0
TutorialCustomer service

Using tool_choice=required for customer service (OpenAI cookbook)

Using Tool Required for Customer Service The ChatCompletion endpoint now includes the ability to specify whether a tool must be called every time, by adding tool_choice='required' as a parameter. This adds an element of determinism to how you build your wrapping application, as you can count on a to

Sourcegithub.com/openai/openai-cookbook/blob/main/examples/Using_tool_required_for_customer_service.ipynb

No credentialsDoes not sendMIT
TutorialCustomer service

Customer service agent with tool use (Claude cookbook)

Creating a Customer Service Agent with Client-Side Tools In this recipe, we'll demonstrate how to create a customer service chatbot using Claude 3 plus client-side tools. The chatbot will be able to look up customer information, retrieve order details, and cancel orders on behalf of the customer. We

Sourcegithub.com/anthropics/claude-cookbooks/blob/main/tool_use/customer_service_agent.ipynb

No credentialsDoes not sendMIT
Awesome listCustomer service

500 AI Agents Projects (includes customer service use cases)

Curated collection of AI agent use cases across industries (incl. Customer Service, E-commerce) with links to open-source implementations, e.g. a LangGraph customer-support agent tutorial.

Sourcegithub.com/ashishpatel26/500-AI-Agents-Projects

No credentialsDoes not sendMIT