salesforcecli/mcp

What it solves

This project provides a way for Large Language Models (LLMs) to interact directly with Salesforce orgs. It bridges the gap between AI models and Salesforce resources, allowing AI to read, manage, and operate Salesforce data and metadata securely.

How it works

It implements the Model Context Protocol (MCP), acting as a server that connects an MCP-compatible client (like Cursor, Claude Code, or Cline) to a Salesforce environment. The server provides a set of specialized tools that the LLM can invoke to perform actions in Salesforce. Users configure the server via a JSON configuration file in their IDE or client, specifying which Salesforce orgs to access and which "toolsets" (groups of tools) to be enabled.

Who it’s for

Salesforce developers and administrators who use AI-powered IDEs or agents to automate development, data management, and org configuration.

Highlights

• Modular Toolsets: Includes over 60 tools organized into categories like metadata, data, users, testing, and devops to prevent overwhelming the LLM's context window.
• Broad Client Support: Compatible with major MCP clients including VS Code Copilot, Claude Code, Cline, Cursor, Trae, Windsurf, and Zed.
• Secure Org Access: Integrates with local Salesforce CLI authorization to ensure secure access to authorized orgs.
• Specialized Migration Tools: Features dedicated tools for migrating Aura components to Lightning Web Components (LWC).
• Dynamic Tool Discovery: An experimental feature that loads tools as needed to improve LLM performance.