Agentforce MCP (STDIO Server)

Agentforce MCP is an MCP (Model Context Protocol) server that enables any MCP-compatible client (like GitHub Copilot or Claude Desktop) to connect directly with Agentforce Agents hosted on Salesforce.

This package bridges your local MCP client with the Salesforce Agentforce platform via STDIO. Once connected, your MCP client can query or perform actions through your configured Salesforce Agent — all in real time.

🚀 Features

• 🔌 Seamless STDIO integration with MCP clients
• 🔐 Salesforce authentication via username, password, and security token
• ⚙️ Agent-based invocation for flexible interaction
• 🧩 Lightweight, CLI-ready NPM package (no local setup required)

🧠 Prerequisites

Before using this package, ensure you have:

• A valid Salesforce account

• An active Agentforce Agent in your org

• Your Salesforce credentials:

  • username
  • password
  • securityToken
  • agentName

• Python, Node.js and npm installed

🧩 GitHub Copilot MCP Configuration

Add the following entry to your mcp.json:

{
  "servers": {
    "AgentforceStdio": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@mohitharsh/agentforcemcp",
        "--username",
        "<your-username>",
        "--password",
        "<your-password>",
        "--securityToken",
        "<your-security-token>",
        "--agentName",
        "<agent-api-name>"
      ]
    }
  },
  "inputs": []
}

💬 Claude Desktop MCP Configuration

Add the following entry to your mcp.json (usually located in the Claude app data directory):

{
  "mcpServers": {
    "AgentforceStdio": {
      "description": "Connect to Agentforce Agent",
      "command": "npx",
      "args": [
        "-y",
        "@mohitharsh/agentforcemcp",
        "--username",
        "<your-username>",
        "--password",
        "<your-password>",
        "--securityToken",
        "<your-security-token>",
        "--agentName",
        "<agent-api-name>"
      ]
    }
  }
}

🧭 Command-Line Arguments

| Argument | Description | Required | Example | | ----------------- | ---------------------------------------------- | -------- | --------------------------- | | --username | Salesforce username for authentication | ✅ | --username [email protected] | | --password | Salesforce password for authentication | ✅ | --password myPass123 | | --securityToken | Salesforce security token for authentication | ✅ | --securityToken aBc123XyZ | | --agentName | Agentforce Agent API Name | ✅ | --agentName Salesforce_Copilot |

⚠️ Note: The Agentforce Agent must be activated and accessible in your Salesforce org.

💡 Practical Example (Claude Desktop)

Below is a real example of the Agentforce MCP server working inside Claude Desktop, showing how Claude communicates with a Salesforce Agent via the MCP server:

User: "get account targets of Genepoint"

The MCP server then sends this query to Salesforce, and the Agentforce Agent responds with structured insights:

Claude:

The Copilot_for_Salesforce agent provided the following information about GenePoint’s account targets:

Account Targets for GenePoint:

1. Reduce delivery time by 30% through route optimization
   • Recommended action: Analyze current delivery routes and implement advanced routing software using real-time data.

2. Launch a new AI product line by the third quarter
   • Recommended action: Establish a clear timeline and allocate resources for timely development and launch.

These strategies focus on enhancing operational efficiency and market competitiveness for GenePoint.

This demonstrates how you can query Salesforce data or agent-driven insights directly from your MCP client without needing to log into Salesforce manually.

🧑‍💻 Author

P. Mohit Harsh AI/ML Engineer • 🔗 GitHub

🪪 License

MIT License © 2025 P. Mohit Harsh