What Is MCP? A Simple Guide To Model Context Protocol for Salesforce Admins

What Is MCP? A Simple Guide to Model Context Protocol for Salesforce Admins

By

One of the challenges of the AI revolution is simply keeping up. Technology evolves quickly: Interfaces get updated, models change in terms of performance, precision, and sometimes even tone and personality.  Completely new functionality is being added to agents at a breakneck speed. For Salesforce Admins, keeping up with new concepts and terminology can definitely be a challenge, especially when new ideas are defined with acronyms or by unfamiliar technological terms.

A great example is Model Context Protocol, or MCP—a new standard designed to let agents talk to each other. Let’s break down what that term means, how the technology works, and what it will mean for admins and their Agentforce implementations.

MCP: A new solution to an old problem

MCP is a protocol drafted by OpenAI and has emerged as the de facto standard for describing how one agent’s model should interact with another’s. For Salesforce, this will allow Agentforce to easily communicate with other third-party agents. Currently this requires custom application development, but in the future it should surface in custom and standard Salesforce functionality.

Ever since computing adopted the client-server model with two very different things trying to talk to each other securely and consistently, the problem of how to make that connection work has existed. How do clients authenticate themselves? How do servers deliver information to the right location? How does the client know what kind of data to expect and respond to it correctly? Now take those questions and realize if someone creates a different kind of server or client, you have to ask them all over again and design an integration that handles different answers.

The Internet Age is full of novel ideas and protocols that solve these kinds of problems. Many application programming interfaces (APIs) include some kind of means of asking for answers to these questions directly, and even import them into client applications. 

The Agentic Age is a similar but different problem compared to traditional internet clients, in that it changes up the kind of questions one would ask an agent versus asking a standard internet application or API.

Context is king

Sound familiar? It should if you’ve been working with any kind of AI. Artificial intelligence will always, always, always be bad at its job if it lacks context. My classic example is the phrase “I need a ticket.” On its own, it lacks the detail needed to complete a task or generate a meaningful response—especially compared to something like “I need a plane ticket from Chicago to San Francisco for Dreamforce 2025.” MCP applies this to an agent that another agent wants to communicate with. MCP also defines the major aspects of an agent you can work with:

  • Host: The main AI platform (e.g., ChatGPT, Gemini)
  • Client: Acts within the host to communicate with MCP servers
  • Server: Exposes agent capabilities and data access
  • Tools: Enable the agent to take actions via your server
  • Resources: Provide structured data and content access
  • Prompts: Define reusable workflows and prompt templates

What do I need to know to make it work?

This is the cool part. Honestly? About as much as you need to turn on your TV. MCP might be a technically fancy name, and has some pretty fancy technical moving parts—but for Agentforce all of that gets handled behind the scenes for you. 

However, the term itself will probably be surfaced in various places as Agentforce uses it for new functionality, so the important part for admins is knowing what it is and a lot less how it works. Agents are going to become more interconnected, and MCP will help make that happen.

What to expect from MCP

As more companies and applications open their functionality up to agentic use cases, MCP will be the protocol that makes it easier for two different agents to work together. Want Agentforce to pull lead suggestions from a LinkedIn agent? MCP. Need it to ask YouTube’s agent for video recommendations based on CRM data? MCP. Looking to check in with Trello on project progress tied to a support case? You guessed it: MCP. Think of all these applications as agents speaking different languages. MCP is your universal translator, making it possible for them to understand each other and work together. 

As the tools for building agents evolve, keep an eye on this space for more details on how MCP will be connected to your agentic solutions. For a more visual overview, check out this video.

Learn more from these resources

Metadata Is Your Admin Blueprint for Building Better Agents

Metadata Is Your Admin Blueprint for Building Better Agents

If the idea of Salesforce Multitenancy is important to understanding Salesforce as a platform, then the logical next step is to consider its use of metadata.  Metadata is fairly ubiquitous in the internet age. We all know it’s important to tag data and content with information and descriptions so that browsers, search engines, and, these […]

READ MORE