Connecting Claude with Salesforce Hosted MCP Servers: A Headless Approach to CRM

Salesforce integrations have traditionally been designed around applications, APIs, middleware, and user interfaces. An external system sends a request, Salesforce processes it, and the result appears in a portal, mobile app, integration layer, or Lightning page.

Generative AI introduces a different kind of consumer: an AI assistant that understands a user’s intent, determines which Salesforce capability to use, retrieves the necessary context, and completes an action.

Connecting Claude with Salesforce Hosted MCP Servers demonstrates this shift. Instead of building a custom integration for every conversational use case, organizations can expose approved Salesforce data and actions to Claude through the Model Context Protocol, or MCP.

This approach also fits naturally into the broader Salesforce Headless 360 vision, where Salesforce capabilities can be accessed through APIs, MCP tools, command-line interfaces, agents, and experiences that do not require users to work directly inside the standard Salesforce interface.

The Model Context Protocol is an open standard for connecting AI applications to external tools and data sources.A traditional API provides endpoints that developers call using predefined request structures. MCP adds a standardized discovery and interaction layer that helps an AI client understand:

Claude can therefore use MCP tools to interact with Salesforce without requiring every Salesforce operation to be manually embedded in a prompt or implemented as a separate custom integration.

Salesforce Hosted MCP Servers provide this tool layer from within the Salesforce platform. An authorized MCP-compatible client can discover and invoke approved Salesforce capabilities while Salesforce continues to enforce authentication, permissions, and platform governance.

Claude is highly effective at understanding natural language, maintaining conversational context, reasoning across information, and selecting tools based on user intent.

Salesforce provides trusted customer data, business processes, automation, security, and transaction management.

Connecting them creates a useful division of responsibilities:

The result is not simply a chatbot that talks about Salesforce. It is a conversational interface capable of securely working with Salesforce.

The exact setup can change as Salesforce and Claude evolve, but the integration generally follows these stages.

Expand the API (Enable OAuth Settings) section and select Enable OAuth.

Enter the following callback URL:

https://claude.ai/api/mcp/auth_callback

The callback URL redirects the OAuth response back to Claude after the user successfully authenticates with Salesforce.

Add the following OAuth scopes:

The refresh-token scope allows Claude to maintain access without requiring the user to authenticate for every request. The MCP API scope grants access specifically to Salesforce-hosted MCP servers.

Only include the scopes required for the integration. Avoid granting broader permissions unless the use case specifically needs them.

In Salesforce Setup, search for MCP servers.

Select the required server and activate it. For example, the sobject-all server provides comprehensive access to Salesforce data. It supports all standard and custom objects in the org, with tools for creating, retrieving, updating, deleting, querying, searching, and navigating object relationships.

Activating the server is essential. Creating the OAuth client alone does not make MCP tools available.

In Claude, open the connector settings and add a custom connector.

Provide:

Use different connector names for each environment, such as:

This reduces the risk of accidentally running a production operation while testing.After saving the connector, complete the Salesforce OAuth login and authorization flow.

Claude can apply different approval policies to the tools exposed through the connector.

Depending on the client configuration, a tool can be:

Prompt – List the accounts in my Salesforce org.

Prompt – Show me the opportunities associated with Magic Account.

Prompt – Mark the Magic Opp opportunity as Closed Won and a Summary.

The Claude integration becomes more strategically important when viewed through the Salesforce Headless 360 model.

Headless does not simply mean removing the user interface. It means separating Salesforce capabilities from any single presentation layer.

In a headless architecture:

Connecting Claude with Salesforce Hosted MCP Servers creates a practical bridge between conversational AI and trusted CRM operations. Instead of building separate custom integrations for every use case, organizations can expose approved Salesforce tools and data through a standardized MCP interface.

Claude can understand a user’s intent, maintain conversational context, and select the appropriate Salesforce capability, while Salesforce continues to enforce authentication, permissions, sharing rules, validation, and business automation.

The real value of this integration is not limited to retrieving records through natural-language prompts. It also enables Claude to help users navigate related data, create records, update information, and complete business tasks without requiring them to work directly inside the Salesforce interface.

For production adoption, organizations should begin with a focused use case, provide only the required MCP tools, apply least-privilege access, and require user approval for sensitive or high-impact actions. With the right governance in place, Salesforce Hosted MCP Servers can turn Claude into a secure and useful conversational interface for Salesforce.

Salesforce Hosted MCP Servers – https://developer.salesforce.com/docs/platform/hosted-mcp-servers/references/reference/servers-reference.html

Claude Code – https://code.claude.com/docs/en/mcp#installing-mcp-servers

Salesforce Hosted MCP Servers (Postman Collection) – https://www.postman.com/salesforce-developers/salesforce-developers/collection/689673309a230a0874d2fb86