What is Model Context Protocol (MCP)?

Model Context Protocol (MCP) is an open, standardized protocol that enables AI assistants to discover and use software capabilities through a common interface. Rather than requiring a separate integration for every AI assistant, an application exposes a single MCP interface that any MCP-compatible AI assistant can use.

MCP is supported by a growing ecosystem of AI assistants, including Claude, Cursor, Codex, Amazon Q, and many others. By standardizing how applications describe their capabilities, MCP simplifies AI integration while improving interoperability across tools and platforms.

Why MCP matters

Traditional software integrations rely on application programming interfaces (APIs), where applications explicitly call predefined functions. This approach works well when software already knows which operation to invoke.

AI assistants operate differently. Before performing a task, they must first determine:

• What capabilities are available?
• Which capability best matches the user’s request?
• What information does each capability require?
• What results each capability produces.

In other words, AI assistants require context in addition to APIs. MCP provides this context by exposing application capabilities in a standardized, AI-friendly format.

As organizations adopt more AI assistants and engineering applications, MCP helps reduce integration complexity by providing a consistent method to discover and access software capabilities.

How MCP works

MCP and APIs

MCP complements existing APIs rather than replacing them. An API defines how to execute a specific operation, while MCP describes all of the operations an application makes available to AI assistants.

The relationship can be summarized as follows:

• API executes a specific operation.
• MCP describes available operations so an AI assistant can discover, understand, and invoke the appropriate one.
• The underlying APIs continue to perform the actual work. MCP provides a standardized description of those capabilities.

MCP clients and MCP servers

An MCP-based architecture consists of two primary components:

• MCP client: Runs within an AI assistant and connects to one or more MCP servers. The client discovers available capabilities, invokes the appropriate tools, and returns the results to the AI assistant.
• MCP server: Runs alongside an application and exposes its capabilities through the standardized MCP interface. The server translates MCP requests into calls to the application’s existing APIs or services.

An MCP server typically exposes three types of capabilities:

• Tools: Actions that the AI assistant can execute.
• Resources: Read-only information that the AI assistant can consult.
• Prompts: Reusable prompt templates that guide domain-specific tasks.

Together, MCP clients and servers enable AI assistants to discover and use application capabilities while leaving the application’s existing APIs responsible for executing the requested operations.

Without/With MCP 

Without MCP, every AI assistant requires a dedicated integration with each application. Every integration must be developed, tested, maintained, and updated independently.

With MCP, an application exposes a single standardized interface that any MCP-compatible AI assistant can use. This helps reduce engineering effort, simplifies maintenance, and makes it easier to support new AI tools as they emerge.

MCP transport methods

MCP defines the protocol used between clients and servers while supporting different methods for transporting messages.

The two primary transport methods are:

• Standard Input/Output (stdio): The MCP server runs locally on the same machine as the AI assistant and communicates through the operating system’s standard input and output streams. This approach is commonly used for desktop applications and development environments.
• Streamable HTTP: The MCP server is accessed via HTTP and can communicate with remote services. This transport is commonly used for cloud services and enterprise deployments where multiple users or AI assistants may access the same server.

Regardless of the transport method, the MCP protocol and the capabilities exposed by the server remain the same.

MCP gateways

Organizations often deploy MCP servers for multiple applications. As the number of servers grows, managing individual connections can become increasingly complex. An MCP Gateway provides a single entry point for AI assistants to access multiple MCP servers.

In addition to simplifying connectivity, an MCP Gateway can provide:

• Centralized discovery of MCP servers.
• Unified authentication and authorization.
• Common identity and permission management.
• Centralized audit logging and monitoring.
• Consistent security and governance policies.
• Simplified deployment and maintenance.

This architecture enables AI assistants to access multiple applications through one standardized interface while preserving the independence of each underlying MCP server.

What MCP does not do

Although MCP standardizes interoperability between AI assistants and software applications, it is not a complete AI platform, and does not:

• Interpret or reason about user requests. This responsibility belongs to the large language model (LLM) and the AI application.
• Provide domain expertise or automatically adapt engineering workflows.
• Manage business logic, workflow orchestration, deterministic execution, memory, or session state.
• Implement enterprise governance measures, including approval workflows, compliance policies, and audit policies.

Its role is to standardize how AI assistants discover and access application capabilities, rather than replacing AI applications or enterprise platforms.

In summary

Model Context Protocol (MCP) is an open standard that enables AI assistants to discover and use software capabilities through a common interface. By complementing existing APIs, MCP simplifies AI integration, improves interoperability, and reduces the need for custom integrations as new AI assistants emerge.