Build and deploy a Support Copilot for Zendesk

Overview

Who this is for: support teams giving their agents a copilot inside the Zendesk ticket sidebar. (For a copilot embedded in your own product for logged-in end users, see In-Product Copilot.)

Support Copilot puts an Inkeep agent right inside the Zendesk ticket sidebar, working alongside your support team. Its edge is the context it can reach that your customer-facing bot never could: your team's internal knowledge — prior resolved threads, internal Slack discussions, and the diagnostics behind past fixes.

When an agent opens a ticket and runs Smart Assist, the copilot will:

1. Read the ticket — the conversation, metadata, and customer details — automatically
2. Search your internal threads for the times your team already solved this exact issue
3. Return structured assistance: a Draft Answer the agent can send, the Relevant Sources behind it (the internal thread and the public article that backed the fix), plus quick replies, clarifying questions, a thread summary, or follow-up tasks as the ticket needs

Everything happens inside the ticket the agent is already working in. No tab-switching, no copy-paste.

Prerequisites

• Access to Inkeep Enterprise — schedule a demo if you don't have it yet
• An internal knowledge source exposed as an MCP server — e.g. your internal Slack support channels or a store of past resolved tickets — that returns prior threads and the public source behind each fix. See Build Custom MCP Servers if you need to create one.
• A Zendesk account, with a Zendesk admin available to approve the app install

Build the Smart Assist agent

Start by building the agent that powers the copilot: a single sub agent that searches your internal threads and helps the support agent reason about the ticket.

Prompt:

You are an AI copilot for [Your Company] that helps the support team make progress on customer tickets.

You're assisting a support agent — they are your user. Every message you receive is from the support agent, never from the customer. The ticket's conversation, metadata, and customer details are provided to you automatically. Focus on where the conversation currently stands.

How to help:
1. When the agent asks you to analyze a ticket — "Smart Assist", "what should I do here", "help me with this" — search your internal threads FIRST. Your team may have resolved this exact issue before; that internal context is knowledge the customer-facing bot never had.
2. Surface what you found together in one response: a Relevant Sources card showing the internal thread AND the public source behind the fix, plus a Draft Answer card with a customer-ready reply that cites only the public source.
3. If the internal thread reveals the root cause (e.g. "not an outage — the bot was never invited to the channel"), state that briefly to the agent before the cards.
4. When the customer hasn't given enough detail to resolve it, ask the questions the agent should send instead of drafting.

Rules:
- Only cite a URL if it appears verbatim in a search result — never invent or guess one. An answer with no source beats one with a fabricated source.
- Never put internal/private links or customer PII in customer-facing content. Internal threads are for the agent only.
- Don't tell the customer to "reach out to support" — they already have.

Connect your internal knowledge

Smart Assist's advantage is the internal context it can reach. Expose that knowledge as an MCP server that gives the agent a search tool — in our example MCP that tool is search_internal_threads, returning a prior internal thread plus the public source that backed the fix. Your MCP may name its tool differently.

Each search result should carry the internal thread, a short resolution summary, and the public source URL that backed the fix — for example:

{
  "thread_content": "Customer's Slack bot wasn't responding when tagged…",
  "resolution": "Not an outage — the bot was never invited to the channel. Fix: /invite @bot, then retry.",
  "public_source_url": "https://slack.com/help/articles/..."
}

Smart Assist links public_source_url in the customer-facing Draft Answer and keeps the internal thread in Relevant Sources, flagged private for the agent.

Your canvas should now show Smart Assist connected to the internal-knowledge MCP:

Add structured assistance

Smart Assist doesn't just reply in plain text — it returns rich cards the agent can act on. You define these as data components: name each one and give it a JSON schema, and the agent fills them in. Create them in the Data Components tab, then attach them to Smart Assist. See Data Components for the full walkthrough.

The two that carry this copilot:

• Draft Answer — a customer-ready reply the agent can send as-is, citing only public sources.
• Relevant Sources — the evidence behind the draft: the internal thread (marked private, for the agent's eyes only) and the public article (safe to link to the customer).

Add more as your team needs them — the demo also ships Quick Replies (one-click acknowledgements), Questions to Ask (clarifying questions when the ticket is thin), a Thread Summary for long threads, and Customer / Agent Tasks for open action items.

Create a Support Copilot app

A Support Copilot app is how you deploy your agent into an external support platform like Zendesk. It connects your agent to the platform and defines the quick actions your team sees in the sidebar.

Go to the Apps tab in the left sidebar and click New App → Support Copilot.

Fill in the app:

• Name: A name your team will recognize (e.g., Support Copilot)
• Default Agent: Select the Support copilot agent you just built
• Platform: Zendesk
• Quick actions: Action buttons shown at the top of the sidebar, organized into named groups. Each button sends a message to the agent on click. Create a group — e.g., ANALYZE — with Smart Assist, Summarize thread, and Draft reply.

Click Create App.

Install in Zendesk

With the app created, install it in Zendesk so it appears in the ticket sidebar. You do this once per Zendesk account.

The in-app install dialog walks you through the same three steps:

Use it in a ticket

Open a ticket in Zendesk and find Inkeep Agents in the sidebar. The copilot greets you with the ticket it's looking at, and your quick actions are ready at the top.

Click Smart Assist and the copilot reads the ticket, searches your internal threads, and — when your team has solved it before — surfaces the prior fix: a brief note on the root cause, the Relevant Sources behind it, and a Draft Answer grounded in the public source, ready to send. Or type your own message to ask it anything about the ticket.

From here you can extend the copilot with more tools as your team needs them — for example, pulling up a customer's logs or checking live integration status. Each one builds on the same agent you just deployed.