Configure MCP tools for AI agents
You can connect external tools to your AI agent using the Model Context Protocol (MCP). MCP tools let your agent take actions in third-party services during a conversation, such as searching a product catalog, checking stock, or building a cart, without leaving the chat.
Who does what
Organization admin: adds the MCP tools in Atlassian Administration.
Customer experience admin or Customer Service Management project admin: connects that source to the agent in Studio.
Before you begin
Your organization admin must connect the MCP server in Atlassian Administration under Connected apps before you can configure tools in your customer experience. For instructions, see Add an external MCP server from Atlassian Administration.
You need Customer Service Management Standard, Premium, or Enterprise.
Step 1: Add MCP tools to your AI agent
To configure which MCP tools your AI agent can use:
In your customer experience, select AI agent from the sidebar, then select Edit in Studio. In Studio, select Tools from the navigation.
Select + Browse ready made tools from an MCP server or Atlassian apps. This list is filtered to show only the tools your AI agent can use.
Select + Browse MCP servers. You'll see the MCP servers your org admin has connected in Atlassian Administration. Servers that haven't been connected at the organization level won't appear here.
Select Add for each tool you want your AI agent to use. You can add tools individually, so you don't have to enable everything a server offers.
Select Save.
Step 2: Test your tools
Before publishing, you can test how the agent uses your selected tools in the test panel.
In the Tools page, select Test to open the test panel in the sidebar.
Send a message to the agent that would trigger a tool — for example, "Show me some t-shirts" or "What's your return policy?"
Check that the agent responds correctly and uses the expected tool.
To see exactly which tool ran, open the conversation review page and select the Test tab. Your test conversations appear there, so you can inspect tool calls before you publish. See Review tool invocations.
Test conversations use your saved tool configuration but do not affect your live agent. You can test as many times as you need before publishing.
Step 3: Publish your agent
When you're happy with your tool configuration, publish your agent to make the changes live. Publishing creates a new version of your agent.
Select Publish.
Review the changes included in this version, then confirm.
Your AI agent is now live with the updated tool configuration. Customers interacting with your agent can trigger these tools through natural conversation.
Tool configuration is versioned. Each time you publish, a new agent version is created. If you need to roll back tool changes, you can restore a previous agent version.
Review tool invocations
The conversation review page shows which tools were called in a conversation, so you can check your tool configuration at any stage — including before your agent is live. Test conversations you run from the test panel (see Step 2: Test your tools) appear on the Test tab straight away, so you can review tool invocations while your agent is still in draft. Conversations with customers appear on the Real tab once your agent is published.
In your customer experience, open the conversation review page.
Select the Test tab to review your own test conversations, or the Real tab to review conversations with customers.
Open a conversation to see the full exchange between the customer and the agent.
Select View steps on a message to see which tool was triggered, what inputs were passed, and what the tool returned.
Use this view to check that tools are being called correctly and to identify any unexpected behaviour. For more on the conversation review page, see Review your AI agent's conversations.
What tools are available
Currently, MCP tools in your customer experience support:
Shopify only — the Shopify Storefront MCP server is the only currently supported integration. The following Shopify tools are available:
Search catalog — finds and browses products in your store
Get product details — retrieves product variants (size, colour), availability, and price
Add to or update cart — builds a cart on behalf of the customer
View cart and checkout — shows cart contents and provides a checkout link
Answer store policy and FAQ questions — retrieves return policies, shipping info, and other store FAQs
No-auth tools only — the currently available tools don't require customers to sign in. Tools that would require customer authentication (such as viewing order history) aren't supported.
Was this helpful?