Agents are becoming ubiquitous tools for end-users to interact with digital services and platforms. Businesses that want to stay relevant in the world where agents and AI applications are the entry point for their discovery and interactions, they must support this new technology by providing a good agent experience (AX). The emerging standard Model Context Protocol (MCP) is a foundational tool for providing a good AX. As an open standard, it is designed to unify how agents are provided context and interact with the capabilities of a service.
The rise of Agent Experience (AX)
The era of Agent Experience (AX) heralds a shift from treating AI purely as a compute engine to embracing it as a holistic partner in workflows. The following AX principles, defined by the Agent Experience community, guide how digital services should empower agents as delegates for human users:
- Human centricity: anchor designs around end-user needs, treating agents as mediums enabling seamless interactions with digital services.
- Agent accessibility: provide agents with parity access to APIs and resources, minimizing barriers and ensuring agents can navigate, discover, and consume services without unnecessary human intervention.
- Contextual alignment: ensure agents receive robust, consistent context about the service, its state, and environment to generate reliable outputs.
- Agent interactivity patterns: define standard communication and interaction patterns (such as authorization flows, voice/tone guidelines, and citation requirements) that agents follow to maintain trust and consistency.
- Differentiate agent interaction: capture and distinguish agent-driven actions in logs, metrics, and audit trails to enable observability, analytics, and security.
These principles establish a foundation for building experiences that are intuitive, reliable, and secure. Realizing these across diverse systems and protocols requires that we lean into the available tools for agentic information delivery and interactivity.
Introducing Model Context Protocol (MCP)
MCP is an open, community-driven specification for giving agents context and access to interact with your digital service. In simple terms, MCP gives your AI assistant a universal remote control, enabling it to do things like fetch databases, manipulate Figma designs, or control music apps, all through a standardized interface.
At its core, MCP provides digital services with ways to:
- Provide the up-to-date, personalized, or dynamic data for agents to incorporate into their end user interactions.
- Provide a means for digital services to explain what functionality a service has and allow them to invoke it.
- Allows digital services to also interact back with the agents themselves to sample data and so forth.
By unifying context packaging, tool discovery, and bi-directional interaction, MCP greatly reduces integration complexity, and accelerates the development of sophisticated agent experiences.
How MCP helps realizing AX
Let’s analyze the impact MCP has on helping us start to achieve the community-driven principles on AX:
1. Human centricity
End users using agents to delegate their interactions with a digital service for whatever reason (e.g. productivity) are looking for reliable means to interact with digital services. MCP simplifies the integration complexity for agents to work with services to achieve the end user’s goals across any agent they want to use. MCP also drastically lowers the bar for giving agents and the end users personalized experiences with little extra work.
2. Agent accessibility
By embedding a uniform context schema and tool definitions in each request, MCP removes bespoke adapters and integration complexity. Agents gain parity access to APIs, context on what to expect, and data sources without manual wiring. This ensures more agents can discover, invoke, and navigate services just as readily as human developers.
3. Contextual alignment
MCP defines a transport-agnostic protocol for conveying environment state, code, document references, whatever is important for agents to know to inform the user This ensures that AI models and service backends work from the same precise context, reducing mismatches and improving the reliability of agent outputs.
5. Differentiate agent interaction
MCP interactions can be made to include metadata, agent identifiers, and so on that mark actions as agent-driven. Downstream systems (logging, monitoring, and audit pipelines) can use these markers to filter, analyze, and secure agent-originated events, empowering observability and accountability.
By adopting emerging protocols like MCP, you ensure every agent interaction is rooted in clear context and able to work reliably with your service. This fundamental alignment between tools, data, and models empowers agents to act decisively and creatively on behalf of users, ushering in a new era where exceptional Agent Experience is not just possible, but very achievable.
Getting started with MCP on Netlify
If you want to start building MCPs on Netlify, check out out these resources:
- A developer guide for building MCPs on Netlify
- Example repos for MCPs on Netlify
For more information on MCP and AX, visit the MCP website for detailed tutorials and the Agent Experience website for guidance on AX and how to approach this discipline.
