buildswithpaul/Frappe_Assistant_Core

Infrastructure that connects LLMs to ERPNext. Frappe Assistant Core works with the Model Context Protocol (MCP) to expose ERPNext functionality to any compatible Language Model

What it solves

Frappe Assistant Core (FAC) allows users to interact with their ERPNext site using natural language via LLMs (like Claude or ChatGPT). It eliminates the need to manually perform repetitive tasks—such as searching for overdue invoices, updating lead statuses, or running revenue reports—by giving the LLM direct, secure access to ERPNext data and workflows.

How it works

FAC acts as a bridge between MCP-ready LLMs and the Frappe/ERPNext ecosystem. It exposes 24 built-in tools for document management, search, reporting, and analytics. The system uses OAuth 2.0 for authentication, ensuring the LLM operates under the specific permissions of the logged-in user.

Key architectural features include:

  • Live Schema Access: It reads metadata directly from the Frappe API on every call, meaning custom fields and permission changes are reflected immediately without needing a sync or index.
  • Skills & Templates: Admins can define "Skills" (instructions on how to handle specific jobs) and "Prompt Templates" (pre-defined starting points for users) to ensure consistency.
  • Extensibility: Developers can add new tools and skills to the assistant via hooks in their own Frappe apps.

Who it’s for

  • ERPNext Users: Teams who want to automate data retrieval and business process management using AI.
  • Frappe Developers: Those looking to extend the AI capabilities of their ERPNext site with custom tools and business logic.
  • Admins: Users who need a secure, audited way to integrate LLMs into their business operations.

Highlights

  • 24 Built-in Tools: Covers CRUD operations, search, report generation, workflow execution, and Python-based analytics.
  • Permission-Aware: Strictly adheres to existing ERPNext user roles and permissions.
  • MCP Compatible: Works with the Model Context Protocol for easy integration with clients like Claude Desktop.
  • Audit Logging: Every interaction is recorded in an Assistant Audit Log for security and compliance.
  • Zero-Sync Schema: No vector indices or snapshot tables; it uses live metadata for real-time accuracy.

Related

  • Project
    xang1234/stock-screener

    A multi-market stock screener and analysis platform that combines technical/fundamental scans with AI-powered theme discovery and research tools.

  • Project
    jd-opensource/JoySafeter

    An AI-native platform for building and orchestrating security agents to automate complex tasks like vulnerability analysis and penetration testing.

  • Project
    holoviz/lumen

    An open-source agent-based framework for chatting with data and RAG, enabling users to generate data pipelines and visualizations via natural language.

  • Project
    microsoft/Build-your-own-copilot-Solution-Accelerator

    A solution accelerator providing reusable architecture and code to build custom enterprise copilots using Azure OpenAI and Azure AI Search for RAG-based data interaction.