Building a Custom Salesforce MCP Server with Apex, Flow, and APIs

Over the years, many of us have built Salesforce capabilities using Apex, Flow, REST services, and platform APIs. Those capabilities were usually consumed by Lightning applications, integration platforms, or external systems.

And then the AI agents introduce another way of using the same business logic.

Instead of building a separate integration for every AI client, Salesforce Hosted MCP Servers can expose selected platform capabilities as tools that an MCP-compatible client can discover and invoke.

In this article, I will walk through how a custom Salesforce MCP server can bring together Apex, Flow, and query-based tools through one governed endpoint.

These are only a few examples. The real value comes from deciding which Salesforce operations should be safely exposed to an agent and how those operations should be governed.

A custom MCP server is a Salesforce-managed endpoint containing a curated collection of tools for a particular persona or workflow.

Instead of giving an AI assistant access to every possible Salesforce capability, an administrator selects only the tools needed for a specific purpose.

Think of the Salesforce org as a large workshop. Apex classes, Flows, queries, and APIs are the available equipment. A custom MCP server creates a smaller tool kit containing only the equipment required for a specific job.

Architecture diagram showing the flow from the user and AI agent through MCP to Salesforce tools, data, and business logic.

This is now an essential section because the official documentation explicitly addresses tool limits.

Salesforce notes that AI clients can struggle to select the correct operation when a server exposes more than a few dozen tools. Custom servers should therefore be focused around a persona or workflow rather than becoming a catalogue of everything available in the org.

A customer-service assistant probably needs case, contact, entitlement, and order-related tools. It does not need metadata deployment, marketing analytics, or user-administration tools.

A focused server improves tool selection, simplifies permissions, reduces risk, and makes testing easier.

Let’s take a deeper look at the Customer Support MCP Server through one practical example and explore the technologies each tool can support.

This server will contain four tools:

A support representative can ask an AI assistant to find a customer’s open issues, create a case, summarize the account, or escalate an urgent problem. The assistant uses the appropriate Salesforce MCP tool rather than attempting to directly manipulate Salesforce data.

Here is a visual representation of the tools and backend technologies that can be used for this use case.

PS – Only auto launched Flows can currently be exposed through this mechanism. The Flow must define its input and output variables. Screen Flows and scheduled Flows are not supported as custom MCP tools. Please review the latest Salesforce documentation on this topic, including key considerations, limitations, implementation details, and recommended best practices.

The salesforce documentation emphasizes that the tool name and description are as important as the underlying implementation because the AI client uses this information to decide which tool to select.

Custom servers can also combine tools from different standard Salesforce servers and organization-specific capabilities under one URL.

In this example, a single endpoint provides access to four specific tools instead of exposing a large set of information that may not be relevant to the task.

https://api.salesforce.com/platform/mcp/v1/custom/CustomerSupportMCPServer

Salesforce also states that custom server configurations can be moved between environments using Metadata API, which should be included in a deployment and lifecycle-management section.

Below are screenshots showing how to create a Salesforce Custom MCP Server and add specific tools to it.

An MCP-compatible AI client can now securely connect to the Salesforce-hosted MCP server using an external client application, OAuth authentication, and the server URL.

Within the same Salesforce-connected Visual Studio Code instance, we can connect to the newly created Custom Remote MCP Server and access all the tools and capabilities it exposes.

Once it’s successfully added. You can view all the tools.

Similarly, the Custom MCP Server can be integrated with Claude using an External Client App, OAuth Client ID, and Client Secret. Refer to my blog for detailed setup instructions:

https://ayaninsights.com/guestblogs/connecting-claude-with-salesforce/

Salesforce Hosted MCP Servers provide platform-managed authentication, authorization, and observability. Salesforce warns against creating an unnecessary proxy around Salesforce REST APIs because doing so can bypass platform controls, reduce telemetry, and introduce additional maintenance.

Each MCP request operates with the permissions of the user who authorized the connection. This includes object access, field-level security, and sharing rules.

Building a custom MCP server inside Salesforce gives AI agents a structured and secure way to use existing Salesforce capabilities. Instead of rebuilding business logic outside the platform, we can expose selected Apex methods, autolaunched Flows, Apex REST services, Aura-enabled methods, and Named Queries as focused MCP tools.

The real value is not simply making more actions available to an AI agent. It is choosing the right tools, defining clear inputs and outputs, applying the correct permissions, and keeping sensitive or high-impact operations governed.

A small persona-based server, such as a Customer Support MCP Server, is a practical place to start because it keeps the use case focused and easier to test.

MCP does not replace Apex, Flow, or APIs. It provides a new way for AI clients to discover and invoke those capabilities while Salesforce continues to manage identity, access, business logic, and data. As agent-based experiences continue to grow, this pattern can help organizations reuse their existing Salesforce investments in a more controlled and scalable way.