MCP server, client and host roles
Key Takeaway
In MCP, the server offers capabilities, the client fetches them, and the host coordinates the exchange. In MCP for Agentforce, Agentforce is both the host and the client, and the external system supplies the server.
Model Context Protocol (MCP) is an open standard for how AI models connect to external systems, including their tools and data. With MCP for Agentforce, agents can use tools from external systems without a custom-built integration.
What MCP is and why it exists
Model Context Protocol (MCP) is an open, vendor-neutral standard for connecting AI models to external tools, data sources, and APIs. Anthropic originally created it.
Anthropic compares MCP to a USB-C port. One standard USB-C port connects many devices. In the same way, MCP is one standard protocol that connects AI models to many data sources and tools. The port stands for MCP, and the devices stand for the data sources and tools.
Without MCP, connecting an agent to an external system means building these parts of the integration yourself:
- API connectivity
- Authentication
- Data access
- Code
MCP acts as a translator between systems, so you don't have to do this manual work yourself.
For example, imagine an agent that needs a shipment's current status from a third-party shipping service during a conversation. With MCP, the agent can get the status from that service. It can then use the status to decide its next step in the conversation.