Skip to main content
Agentforce (Salesforce’s generative AI platform) can call external MCP servers. Host your tools on Heroku and register the manifest in Salesforce so custom agents can invoke them alongside CRM data.

​
Prerequisites

  • Access to Agentforce with the Model Context Protocol feature enabled.
  • A Heroku MCP server with a public manifest (for example https://your-app.herokuapp.com/.well-known/mcp.json).
  • Salesforce credentials with permission to manage Connected Apps.

​
Register the manifest

  1. From Salesforce Setup, open Agentforce → Tools → MCP Integrations.
  2. Click New MCP Server and supply:
    • Name: descriptive label (for example Heroku Deploy Tools).
    • Manifest URL: the Heroku /.well-known/mcp.json endpoint.
    • Authentication: API key or OAuth details, if your server requires auth.
  3. Save the integration. Salesforce fetches the manifest and validates tool definitions.

​
Attach to an agent

  1. Open the Agentforce builder for the agent you want to enhance.
  2. In Tools, add the MCP integration you created (for example, Heroku Deploy Tools).
  3. Configure prompts or guardrails so the agent knows when to call each tool.
  4. Test the agent in the preview console; Heroku logs show tool invocations in real time.

​
Best practices

  • Use Salesforce Shield or Named Credentials to store sensitive auth tokens securely.
  • Keep tool responses structured—Agentforce agents can map JSON output into flows or records.
  • Monitor invocation metrics in Salesforce and Heroku to maintain reliability SLA targets.
  • Combine MCP tools with Agentforce’s CRM grounding so assistants can reason over Salesforce data while executing Heroku automation.
See the MCP integration cookbook for deployment tips that apply to Agentforce, Claude, and ChatGPT.