Unlocking new possibilities: Revolutionizing service with the Zendesk AI MCP client

Customer support interactions heavily depend on personalized, real-time information, like order statuses, account details, interaction history, and more. But today, connecting this data with AI assistants requires APIs, an experienced developer (or two), and lengthy lead times.

For example, to renew a customer’s streaming subscription, an AI agent would need their account ID, subscription plan, and billing information – details that might be stored in multiple CRM or payment systems.

Retrieving this data requires multiple integrations, but most service teams lack the technical expertise to do this. This forces leaders to budget time and developers to understand the APIs and authentication methods before they’re able to build the custom integration. Even if there’s a pre-built integration to one system, it might not easily connect the fields the AI agent needs.

You can make it work, if you have the time and money to spend. But there’s a better way.

At Zendesk, our mission is to enhance customer experiences with smarter, easier, more adaptable AI solutions. That’s why we’re bringing Model Context Protocol (MCP) into the Zendesk Resolution Platform. MCP’s open-source model offers a simple, standardized path towards integrating AI with business data. With MCP, Zendesk’s AI Agents and Copilots get the relevant, real-time data they need to go beyond isolated language models, improve AI accuracy, and drive better outcomes for agents and customers.

What is a Model Context Protocol (MCP)?

AI assistants can connect to external tools with APIs, but this often requires custom, one-off integrations with limited extensibility and scalability. The Model Context Protocol (MCP), introduced by Anthropic, changes this. It’s a standard, open framework that gives AI assistants a simple, unified way to access and interact with external data.

Think of MCP as your personal concierge. To integrate a service and CRM system today, you would need to sift through developer documents and figure out how to connect your data. Each data point (ex. Account ID, subscription plan, etc.) might also require a separate API integration. While APIs are flexible, they evolve over time, requiring admins to keep an eye out for new changes and capabilities. This is not the “set it and forget it” era.

But what if you had a personal concierge that did all this work for you? Unlike APIs, MCP integrations only need to be set up once. And once a connection is made, AI assistants can use MCP to access new and existing capabilities within a system – no additional set up required. This streamlined process reduces the amount of time and resources spent on custom integrations. Best of all, it also helps AI agents and copilots reach smarter, faster decisions without losing context.

How does MCP work?

MCP uses a “client-server” model to follow a simple, four-step process:

1. Customer request: The AI agent, or copilot, identifies information needed to complete a customer’s request.
2. MCP request: The AI assistant sends a request to an MCP client, asking for the necessary data or requesting an action to be completed externally.
3. Data retrieval: The MCP client relays the request to an external system (ex. a CRM) through their MCP server.
4. Server response: The server returns information to the client, who delivers it back to the AI assistant to inform their next response.

In this case, the MCP clients act as a bridge between host platforms, like Zendesk, and external servers, allowing AI assistants to send and receive information from a single interface. This modular approach ensures AI assistants are always context-aware without heavy developer involvement, improving both scalability and maintainability.

How is MCP ushering in the next era of AI agents?

Unlike proprietary frameworks, which can be siloed and difficult to scale, MCP’s open-source model is continuously evolving. And with its standardized integrations, AI agents can effortlessly retrieve, update, and interact with data from any system that has an MCP server. This transforms AI agents into dynamic resolution-engines that can autonomously personalize conversations with fresh, relevant information.

This offers endless opportunities for both AI agents and admins. More specifically:

• Higher operational efficiency: Admins can set up an MCP client once and use it to continuously access hundreds of actions and data points. This gives them greater control over their AI assistants, and frees up capacity for admins and developers alike.
• Autonomously resolve more problems, faster: MCP gives AI agents access to a larger, broader set of actions, allowing them to perform tasks that traditionally required human intervention. Now, MCP-powered agents can instantly access the tools they need, boosting AI accuracy, reducing agent workload, and accelerating time to resolution.
• Accelerated service innovation: Whenever MCP clients and servers connect, servers can promote new actions to AI agents. Over time, agents become capable of resolving issues in creative ways, within guardrails. As AI agents evolve, admins can spend less time building integrations and procedures, and more time innovating best-in-class customer experiences.

For example, a customer on the market for a specific coffee blend might ask an AI agent if it’s available for immediate shipping. Traditionally, an AI agent would offer general shipping information, but with MCP, the AI agent can check the inventory management system, confirm stock availability, and suggest alternative products if the item is out of stock. It can also set up the shipment, and set up package shipping tracking. Not only does this instantly resolve the customer’s issue, but it offers accurate information that leads to a revenue-generating outcome — all without human intervention.

The future of MCPs in AI-first service

As leaders in AI-powered service, Zendesk believes MCP is foundational to building smarter, more connected AI assistants. Since Anthropic first launched the MCP standard, there are hundreds of MCP servers with more coming online every day. That is why we’ll soon be releasing a new Zendesk MCP Client for AI Agents and Copilot.

With this release, Zendesk admins can create custom MCP client actions with MCP servers and use them with Copilot, AI Agents, and Action Builder workflows. For example, to automate an integrated CRM workflow, admins simply create a Custom Action, select the CRM server they want to connect to, and select an action or capability from a list of available tools. From there, they can reference the action in an action flow. Every time the flow is triggered, the integration is processed through the Zendesk MCP Client, which uses the CRM’s MCP server to fetch the necessary data and return it to the AI Agent or Copilot to inform its responses.

*Zendesk MCP Client and Action Builder. Imagery may differ from the final product.

To be first in line for early access, customers can join the MCP Client EAP waitlist. And to learn more about Zendesk’s latest innovations, register for the 2025 AI Summit.