# Working with a helpdesk (Pylon, Zendesk)

> Mirrored helpdesk tickets are regular issues: the same tools triage them, and replies deliver into Pylon or Zendesk.

Product: Clad MCP
Source: https://docs.useclad.ai/mcp/helpdesk

---

If the workspace has Pylon or Zendesk connected, tickets from the helpdesk are
mirrored into Clad as regular issues. Nothing changes for your agent: the same
issue tools work on them, no helpdesk-specific tools or scopes needed.

- **Read and triage** — `search_issues`, `get_issue`, `set_issue_status`,
  `set_issue_priority`, `assign_issue`, `set_issue_tags` all work on mirrored
  tickets. New customer messages arrive via the helpdesk's webhook within
  seconds.
- **Reply** — `reply_to_issue` delivers the reply **into the helpdesk itself**,
  as the workspace's connected support seat. The customer sees it in Pylon or
  Zendesk (and by email if that's the ticket's channel). The result's
  `delivery` field is `pylon` or `zendesk` on success.
- **Internal notes** — `reply_to_issue` with `isInternal: true` stays in Clad
  and is never written to the helpdesk.

Two things to know before replying:

1. **The connection must be in Assist mode.** In Observe mode Clad never writes
   to the helpdesk; the tool returns a `FORBIDDEN` error saying so. An admin
   switches modes in Settings → Integrations — this is enforced server-side, so
   there is no way around it from the agent.
2. **Sends are idempotent.** Retrying the same reply to the same customer
   message never double-posts; the duplicate returns
   `delivery: pylon_deduplicated` / `zendesk_deduplicated`.

### How this relates to Clad's built-in agents

Clad also ships its own agents for helpdesk workspaces, and they run in two
places: the Clad app, and the **Clad panel embedded inside the Zendesk ticket
sidebar / Pylon issue view**. The panel is where human agents run code and
billing investigations on the ticket they're viewing, turn findings into a
suggested reply, approve it, and file tracker tickets — all without leaving
the helpdesk. Clad also auto-drafts suggested replies on new customer
messages, which teammates approve in Clad or in the panel.

The same workflow is exposed to external agents too — see
[Investigations, drafts & escalation](#investigations-drafts--escalation) for
running investigations, reviewing and approving drafts, and filing tracker
tickets over MCP. Customer messages and approved replies all land in the
issue's message thread, so `get_issue` gives an external agent the same
conversation your team sees.

That's the whole model: connect your helpdesk in Clad once (Settings →
Integrations, or the guided setup under Agents), connect your agent to this MCP
server, and the agent can work the helpdesk queue end to end — triage,
investigate, draft, reply, escalate — without a helpdesk API integration of
its own.