A conversational assistant powered by LLMs to explore and interact with your Salesforce organization using natural language prompts. It turns complex Salesforce tasks into smooth, human-like conversations directly from a chat interface.

The Salesforce MCP Assistant uses OpenRouter LLMs (e.g., GPT-4.1 mini) to intelligently interpret user inputs, identify the right Metadata Control Plane (MCP) tool, execute queries or administrative actions via Salesforce APIs, and return structured, user-friendly summaries.

Clickable URLs in responses provide direct access to Salesforce records in Lightning UI, making navigation seamless. Built-in session memory allows the assistant to maintain context across turns for coherent, personalized conversations.

This assistant simplifies Salesforce data exploration by letting users query, list, and retrieve object metadata without writing SOQL. It’s designed for both technical and non-technical users who want to access Salesforce data conversationally.

Key capabilities include:

• Listing records from Salesforce objects.
• Counting records.
• Retrieving field metadata.
• Generating clickable record links.
• Summarizing query results in plain English.

• Python 3.10+
• Salesforce Org credentials (username, password, security token, domain, instance URL)
• OpenRouter API Key (for LLM responses)
• Hosting platform like Railway (or local execution)

Create a .env file (or copy from .env.example) with:

SF_USERNAME=your_salesforce_username
SF_PASSWORD=your_salesforce_password
SF_SECURITY_TOKEN=your_salesforce_security_token
SF_DOMAIN=login
SF_INSTANCE_URL=https://your-instance.lightning.force.com
OPENROUTER_API_KEY=your_openrouter_api_key

1. Clone the repository

   git clone https://github.com/sab110/MCP_Salesforce.git
   cd MCP_Salesforce

2. Set up a virtual environment and install dependencies

   python3 -m venv venv
   source venv/bin/activate
   pip install -r requirements.txt

3. Configure environment variables
   Update .env with your Salesforce credentials and OpenRouter key.

4. Launch the assistant

   streamlit run streamlit_app.py --server.port 8502

• salesforcemcp/ contains MCP tool logic:
  • implementations.py: Tool functions.
  • definitions.py: Tool metadata.
  • sfdc_client.py: Salesforce connection and query execution.
• streamlit_app.py drives the chat interface.
• Tools are listed dynamically using metadata, dispatched via function maps, and summarized via LLMs.

• Chat-style UI with threaded conversations.
• Realtime assistant responses with clickable Salesforce record URLs.
• LLM summaries for raw tool outputs.
• Console logs for debugging.

Deploy on Railway or similar platforms:

1. Connect your GitHub repo to a Railway project.
2. Configure environment variables (SF_USERNAME, SF_PASSWORD, etc.).
3. In Railway deployment settings, set the start command:

   streamlit run streamlit_app.py --server.port=8081 --server.headless=true

4. Ensure Railway networking uses port 8081.
5. Enable "Wait for CI" in Railway’s GitHub integration for auto-redeploy on every push.

To adapt the assistant for specific clients:

• Update .env with client-specific Salesforce credentials and OpenRouter API key.

src/
├── assets/
├── salesforcemcp/
│   ├── __init__.py
│   ├── definitions.py
│   ├── implementations.py
│   ├── sfdc_client.py
│   └── server.py
├── streamlit_app.py
├── requirements.txt
├── .env.example
└── pyproject.toml

Built and customized by request for client-specific Salesforce assistant usage.