This YouTube demo shows how AgentForce uses the Model Context Protocol (MCP) to enable AI agents to interoperate with external tools—specifically Salesforce, PayPal, and Slack—without bespoke integrations. The presenter walks through AgentForce Builder: adding a certified MCP server (from Agent Exchange), authenticating PayPal, controlling allowed MCP actions, and wrapping actions with natural-language business policies for governance and security.
Chapters
In this demo, I'll show you how MCP will bring enhanced interoperability to AgentForce. This is an AI agent I built with AgentForce that helps me work, and I want AgentForce to retrieve an order from Salesforce and generate an invoice with PayPal. And I want this to happen seamlessly across these systems because this interoperability is key to enabling AgentForce to do more and more on our behalf. And that is why there is so much excitement about MCP or Model Context Protocol, which is an emerging open standard for how AI models connect with external tools, systems and data. Just Imagine you want AgentForce to use several external tools without MCP. You'd need to create complicated point to point integrations for each new tool, which means you're handling all the API, connectivity, authentication, data access code, and so much more. And MCP solves this problem by acting as a universal translator for these external tools. But there is nuance to be aware of between AgentForce and these tools is a built in MCP client which connects to one or more MCP servers that act as translators that you'll need to trust with your authentication keys. And behind those MCP servers are the tools that are being made available to AgentForce only, some of which you might actually want AgentForce to use. And you'll want to add instructions actions to ensure Agent Force is using these external tools in accordance with your business policies. The bottom line, MCP is great, but only when combined with security, governance and trust. So how will Agent Force help you bring the benefits of McP to your AI agents? Well, here I am in AgentForce Builder, where admins and developers build AI agents, and I'm building the agent you saw earlier. But when I ask AgentForce to generate a PayPal invoice, what you'll see is AgentForce can't use PayPal yet, and soon you'll be able to use MCP and agentforce to fix this by adding a new MCP action from mulesoft, heroku or Agent Exchange. Agent Exchange is where you'll find certified MCP servers from Salesforce's partner ecosystem that have been reviewed and validated, including a PayPal MCP server. On this screen, I could review this server's details, add it to my agent, and then authenticate into PayPal to manage this authorization. And notice that Agent Force will provide full control over which MCP actions the agent is allowed to use, which helps prevent unexpected behavior. And that's it. When I click confirm. I've now integrated AgentForce with PayPal without needing to build any custom integrations, but I am not finished yet. I want to wrap these external MCP actions with some extra guardrails and business policies, which I can do by creating an instruction in natural language, not code to instruct AgentForce to send a Slack message about invoices above a certain amount. And now I'm ready to test it by asking Agent Force once again to generate a PayPal invoice for an order. And we'll see a few things. First, Agent Force is extending interoperability across multiple systems Salesforce, PayPal, and Slack. Second, Agentforce is using its built in MCP client to take action on PayPal. And third, AgentForce is wrapping our business policies around these MCP actions. So now you've seen how MCP will bring enhanced interoperability to AgentForce. To learn more, visit agentforce.com thanks for watching.
