MAC Project Demo

This project demonstrates how to integrate some of the MuleSoft AI Chain project's modules and connectors for your Mule projects.

In this Project

Agentforce

Connect to your Agentforce Agent via API.

• Conversation
  • Start
  • Continue
  • End

Einstein AI

Use the Einstein AI trust layer to make use of different models from Einstein.

• Agent
  • Define Prompt Template
• Chat
  • Answer Prompt
  • Generate From Messages
• Embeddings
  • Adhoc File Query
  • Generate From File
  • Generate From Text
• RAG
  • Adhoc Load Document

Tip

Agent, Chat and RAG make use of LLMs to give an answer in natural language, whereas Embeddings will give numeric responses since no LLM is being used.

Tools - Use AI Service operation is not included in this project.

MuleSoft AI Chain

Build a unification layer for LLMs, vector stores (in-memory), and AI Services.

• Agent
  • Define Prompt Template
• Chat
  • Answer Prompt
  • Answer Prompt with Memory
• Embeddings
  • New Store
  • Add Document to Store
  • Add Folder To Store
  • Query From Store
  • Get Info from Store
• Image Generation
  • Read
  • Generate
• RAG
  • Load Document
• Sentiment Analysis
  • Analyzer
• Tools
  • Use AI Service

MuleSoft Inference

Long-term successor of MuleSoft AI Chain Connector

• Tools
  • Use Native Templates
• Toxicity
  • Detection by

MuleSoft WebCrawler

A connector toolkit that provide web crawling capabilities to extract data from web pages.

• Crawl
  • Website (Full Scan)
  • Get Sitemap
• Page
  • Download Document
  • Download Image
  • Get Content
  • Get Insights
  • Get Meta Tags
• Search
  • Google

MuleSoft Whisperer

Support Speech-to-text and text-to-speech to enhance other MuleSoft AI Chain connectors.

• Speech to Text
• Text to Speech

Prerequisites

• Salesforce org - Create a new dev account here
• Anypoint Platform account - Create a free trial account here
• MuleSoft IDE - Anypoint Studio or Anypoint Code Builder
• Mule Runtime 4.9.x
• Maven 3.x
• Java 17
• OpenAI API key - Create OpenAI account here, generate OpenAI API key.

Project Structure

├── api-spec/         # API Spec files to be published in your Exchange
├── src/                  # Source code directory
├── postman/              # Postman collections + envs for testing
├── mule-artifact.json    # Mule runtime and Java versions
└── pom.xml               # Maven dependencies

Salesforce Configuration (for Agentforce/Einstein AI)

1. Salesforce Connected App

Note

Based on these instructions.

1. Open your Salesforce org and to go Setup.
2. From setup, use Quick Find to search for App Manager.
3. Click New Connected App and then select Create a Connected App.
4. Specify a Connected App Name like mac_project.
5. Add your contact email. Make sure it's a real email, you'll need it later.
6. Check Enable OAuth Settings ✅
7. Callback URL: https://login.salesforce.com
8. Selected OAuth Scopes. Do NOT select more scopes or you'll get a scope error:
   • Access chatbot services (chatbot_api)
   • Access the Salesforce API Platform (sfap_api)
   • Manage user data via APIs (api)
   • Perform requests at any time (refresh_token, offline_access)
9. Deselect/select the following settings:
   • Require Proof Key for Code Exchange (PKCE) Extension for Supported Authorization Flows ◻️
   • Require Secret for Web Server Flow ◻️
   • Require Secret for Refresh Token Flow ◻️
   • Enable Client Credentials Flow ✅
   • Enable Authorization Code and Credentials Flow ◻️
   • Enable Token Exchange Flow ◻️
   • Enable Refresh Token Rotation ◻️
   • Issue JSON Web Token (JWT)-based access tokens for named users ✅
   • Introspect All Tokens ◻️
10. Save the app and click Continue.
11. Once you're in the Manage Connected Apps page, click Manage.
12. Click Edit Policies.
    • Permitted Users: All users may self-authorize
    • Enable Single Logout ◻️
    • Run As <select your user, in my case, Alex Martinez>
    • Issue JSON Web Token (JWT)-based access tokens ✅
    • Token Timeout: 30 Minutes
13. Go back to the connected app and click Manage Consumer Details.
14. Add your verification code from your email.
15. Click generate and then apply.
16. Copy the consumer key/secret from the top. You will use these in MuleSoft.

2. Agentforce Setup

1. In Setup, go to Einstein Setup and turn on Einstein.
2. Wait a few seconds and refresh the site. You might not see the following options if you don't refresh.
3. Look for Agentforce Agents and turn on Agentforce.
4. Enable the Agentforce default Agent.
5. In Agentforce Agents, click on New Agent.
6. Create from a Template > Agentforce Service Agent
7. Remove all the topics to start from scratch
   • (optional) Add the Off Topic topic with the information from off-topic.md
8. Add the following details:
   • Name: Agentforce MuleSoft Agent
   • API Name: Agentforce_MuleSoft_Agent
   • Description: Use custom topics and actions to connect to your MuleSoft APIs.
   • Role: An AI customer service agent who connects to MuleSoft APIs for custom functionality.
   • Company: This company is used for demo purposes.
9. Don't add Data Cloud resources for this example.
10. Once the agent is created and you have it open, click Connections.
11. Click on Add to add a new connection.
12. Select API, add an integration name like MAC Project Integration and select the connected app you previously created (mac_project).
13. Once you add the connection, make sure to click on Activate at the top-right of the screen.

Important

The Mule API cannot call Agents that are deactivated. Make sure that it's Active in Salesforce. BUT anytime that you need to modify the agent, you have to deactivate it first.

Mule Configuration

The project uses secure configuration properties for sensitive information. Make sure to configure the following:

• Create and publish an API Spec based on the files in api-spec/

  • Make sure to match your own org ID and RAML version in pom.xml and mac-api.xml

• OPENAI_API_KEY environment variable in your OS (if running locally)

• properties.yaml file under src/main/resources

• secure-properties.yaml file under src/main/resources

  • Configure your environment and Salesforce credentials:

    domain: YOUR_SALESFORCE_ORG_DOMAIN
    clientId: YOUR_CONNECTED_APP_CLIENT_ID
    clientSecret: YOUR_CONNECTED_APP_CLIENT_SECRET

    Your Client ID / Secret is the same from your Connected App's Consumer Key / Secret. You can obtain your Salesforce Org's domain from Salesforce Setup > My Domain > Current My Domain URL. For example: orgfarm-6acf46e05z-dev-ed.develop.my.salesforce.com

• Anypoint Platform credentials in your own .m2/settings.xml file

  <servers>
      <server>
          <username>your_AP_username</username>
          <password>your_AP_password</password>
          <id>anypoint-exchange-v3</id>
      </server>
  </servers>

Running the project

The project includes Postman collections and environments for testing the APIs. You can find them in the postman/ directory.

Troubleshooting

• Authentication errors: Double-check your Client ID, Client Secret, and Connected App settings.
• Scope errors: Make sure the scopes in your Connected App are set up correctly. If you add more/less scopes than the mentioned in the settings here, you might get scope errors.
• Network issues: Ensure your Mule server can reach Salesforce.

References

• Salesforce OAuth Scopes Documentation
• Salesforce Agent API - Get Started
• Agentforce Supported Operations - Conversation
• MuleSoft Whisperer Getting Started
• MuleSoft WebCrawler Getting Started
• MuleSoft AI Chain Getting Started
• MuleSoft Inference Getting Started