Shield Glossary

MCP (Model Context Protocol)

Back To Glossary Page

What is Model Context Protocol (MCP)?

Model Context Protocol (MCP) is an open standard that enables AI applications, assistants, and agents to connect with external data sources, systems, and tools through a common interface. It provides a standardized way for an AI model to access relevant business context and interact with approved capabilities rather than relying solely on the information contained within the model itself.

MCP can be thought of as a connectivity layer between AI applications and enterprise systems. Instead of building a separate custom integration for every combination of AI tool and data source, organizations can expose approved resources and capabilities through MCP servers. MCP-compatible AI applications can then discover and use those resources according to the access controls and permissions applied to them.

For enterprises, this becomes particularly important as AI moves beyond standalone chat applications. Employees increasingly want AI assistants to work with information held across business systems, databases, applications, and internal platforms. MCP provides a standardized framework for making those connections.

How MCP Works

MCP uses a client-server architecture. An AI application can act as an MCP client and connect to one or more MCP servers. Those servers provide controlled access to specific data, tools, or capabilities.

The basic components include:

  • MCP hosts — AI applications or environments where users interact with AI models and agents
  • MCP clients — Components within the host that establish and manage connections with MCP servers
  • MCP servers — Services that expose approved data, tools, or capabilities to compatible AI applications
  • Resources — Information that an MCP server makes available as context, such as documents, records, application data, or structured datasets
  • Tools — Functions that an AI application can invoke through the MCP server, such as running a query or performing an approved action
  • Prompts — Reusable instructions or workflows that an MCP server can make available to AI applications

Together, these components allow an AI application to discover what information and capabilities are available and use them when responding to a user request.

What MCP Enables

The practical value of MCP is that it allows AI systems to work with information that exists outside the model.

An MCP connection can enable an AI assistant to:

  • Retrieve enterprise data — Access relevant information from databases, applications, repositories, and internal platforms
  • Search business systems — Find records or information without requiring users to manually copy data into an AI application
  • Use specialized tools — Invoke approved functions exposed by an organization’s systems
  • Combine multiple sources of context — Bring information from different enterprise applications into the same AI workflow
  • Support AI agents — Give autonomous or semi-autonomous agents access to the tools and information required to complete multi-step tasks
  • Maintain controlled access — Allow organizations to determine which resources and capabilities are exposed to AI applications

This makes MCP particularly relevant to enterprise AI, where the usefulness of an AI assistant often depends on its ability to work with proprietary information and operational systems.

MCP and Enterprise AI

Large language models are powerful reasoning and language systems, but they do not automatically have access to an organization’s internal information.

Without integrations, an AI assistant may know how to analyze a compliance alert, for example, while having no ability to retrieve the firm’s actual alerts.

MCP provides a standardized mechanism for connecting those two layers.

A compliance team using an MCP-enabled AI environment could potentially ask questions such as:

  • What are our highest-risk alerts this week?
  • What is the breakdown of our alert queue by status?
  • Which risk behaviors are generating the most alerts?
  • Which investigators currently have the largest workloads?
  • Show the trend in alerts associated with a particular risk behavior.

The AI application interprets the request, while the connected MCP server provides access to the relevant enterprise information or functionality.

MCP and AI Agents

MCP becomes especially important as organizations adopt AI agents.

Traditional AI assistants primarily generate responses. Agents can perform more complex workflows that involve retrieving information, using tools, making decisions within defined parameters, and completing sequences of tasks.

For an agent to operate effectively inside an enterprise, it needs reliable access to the systems where work actually happens.

MCP can provide that connection.

An investigation agent, for example, could use MCP-accessible tools to retrieve an alert, gather related communications, examine relevant contextual information, and present its findings to an investigator. Different MCP servers can expose different capabilities, allowing organizations to build AI workflows across multiple systems without requiring every connection to use a completely different integration architecture.

MCP and Governance

Connecting AI applications to enterprise systems introduces an important governance question: what should the AI be allowed to access and do?

MCP provides the connection framework, but the existence of an MCP connection does not automatically make an AI workflow secure or compliant. Organizations still need appropriate authentication, authorization, access controls, logging, data governance, and oversight.

For regulated organizations, important considerations include:

  • Authentication — Verifying the identity of users and applications requesting access
  • Authorization — Ensuring users and AI applications can access only permitted information and capabilities
  • Data controls — Determining what information can be exposed through an MCP server
  • Auditability — Maintaining records of interactions and actions where required
  • System-of-record controls — Ensuring authoritative enterprise systems remain the source of governed business information
  • AI governance — Establishing policies around how AI applications and agents can use enterprise data and tools

MCP standardizes connectivity. Organizations remain responsible for governing what happens through that connection.

Why MCP Matters for Financial Services

Financial institutions operate across large numbers of specialized systems containing sensitive and regulated information. Compliance, surveillance, legal, risk, trading, communications, and operational data may all exist in separate environments.

As firms introduce enterprise AI assistants, those systems become potential sources of context for AI-driven workflows.

MCP creates a standardized way to expose selected capabilities to those AI environments while allowing the underlying enterprise platforms to remain authoritative systems of record.

In communications compliance, for example, an MCP server could allow an authorized user to access surveillance intelligence from within an enterprise AI environment rather than requiring the user to manually move between applications.

The opportunity is significant, but so is the governance requirement. AI access to regulated information must remain subject to the firm’s existing security, permissions, retention, supervision, and compliance frameworks.

MCP vs. APIs

MCP does not eliminate APIs. In many implementations, APIs remain part of the underlying infrastructure used to retrieve information or execute actions.

The difference is the interface presented to AI applications.

Traditional APIs are generally designed as application-to-application integrations with implementation-specific endpoints, authentication methods, schemas, and documentation. An AI integration therefore often requires custom development for each system.

MCP provides a common protocol designed specifically for connecting AI applications with external context and tools.

An MCP server can sit in front of existing APIs and expose selected capabilities in a standardized format that MCP-compatible AI applications can discover and use.

This makes MCP complementary to existing enterprise integration infrastructure rather than a wholesale replacement for it.

MCP Servers and MCP Clients

The distinction between an MCP server and MCP client is important.

An MCP server exposes information or capabilities. An enterprise software provider, for example, might operate an MCP server that provides governed access to data held within its platform.

An MCP client connects to that server on behalf of an AI application.

This architecture allows an organization to connect an AI environment to multiple MCP servers, potentially giving the AI access to different enterprise systems through the same protocol.

The result is a more interoperable ecosystem in which AI applications can work with external systems without every connection requiring a completely bespoke integration.

Synonyms and Related Terms

  • Model Context Protocol
  • MCP server
  • MCP client
  • AI connector
  • AI integration
  • Enterprise AI integration
  • Agentic AI
  • AI agents
  • AI tools
  • Tool calling
  • Enterprise AI governance
  • AI interoperability

Related Terms