AI + ERPNext Integration Platform
Frappe Assistant Core: The LLM Integration Layer for ERPNext
A Techunison-style case study on how Frappe Assistant Core turns ERPNext into an AI-connected operating system by exposing business functionality to any MCP-compatible language model through secure, standardized tools and protocols.
This is not just an AI connector. It is an enterprise execution layer that lets LLMs interact with ERPNext safely, audibly, and at production scale.
Model Strategy
LLM-Agnostic
Compatible with Claude, GPT, custom models, and any MCP-enabled system.
Tool Coverage
20+ Built-In Tools
CRUD, search, reporting, analytics, Python execution, and visualization workflows
Enterprise Layer
OAuth + RBAC
ERPNext permissions, audit logging, role-based access, OAuth 2.0 and OIDC
Extensibility
Plugin Architecture
Custom business tools, external integrations, and industry-specific logic.
01
The challenge
LLMs can reason in natural language, but enterprise systems remain isolated behind brittle APIs, custom scripts, and one-off integrations.
02
The opportunity
ERP users want AI that can do real work: search records, create documents, run reports, and follow permissions.
03
The blocker
Without a standard protocol, every model integration becomes a maintenance problem and every vendor change creates new lock-in.
04
The answer
Frappe Assistant Core uses MCP to expose ERPNext capabilities to AI systems through a reusable, secure, standards-driven integration layer.
Platform Positioning
What Frappe Assistant Core Really Is
Frappe Assistant Core is infrastructure that connects large language models to ERPNext. Instead of building separate integrations for each AI provider, it creates one standardized access layer using the Model Context Protocol. That means ERPNext functionality can be exposed once and then consumed by Claude, GPT, custom internal agents, or any other MCP-compatible client.
That shifts the conversation from “AI add-on” to “AI operating layer.” The system is designed for real enterprise use: secure authentication, ERPNext-native permissions, auditability, plugin extensibility, and built-in tools that map directly to operational work.
Why this matters to enterprises
No model lock-in: change LLMs without rebuilding the ERP bridge.
One integration pattern: standardized MCP interface instead of provider-specific orchestration.
Governed access: authentication, role-based access, and audit logs are part of the design.
Business-ready extensibility: custom plugins can encode industry logic, workflows, and integrations.
Architecture Overview
Core Platform Components
MCP Server
Protocol handler that exposes ERPNext functions as standardized tools for AI clients.
OAuth / OIDC
Dynamic client registration, PKCE support, discovery endpoints, and secure authorization flows.
Tool Layer
21 built-in tools for documents, search, analytics, reporting, Python execution, and visualization.
Plugin Framework
Extend core behavior with custom business logic, external integrations, and vertical-specific tools.
ERPNext Security
Native permissions integration, audit trail, and enterprise controls around every model action.
Built-In Capabilities
21 Tools That Turn AI Into an ERP Operator
Frappe Assistant Core does not stop at chat. It exposes useful enterprise actions and data flows directly into AI clients.
Document CRUD
Create, read, update, and work with ERPNext documents through structured tools.
Search
Natural language and tool-based search across ERP entities and business records.
Reporting
Run operational and management reports from the AI interface.
Analytics
Summaries, comparisons, and metric-driven insights on ERP data.
Visualization
Generate tables and charts for business users from tool outputs.
Python Execution
Support advanced calculations and logic-heavy workflows where needed.
User Workflows
How Business Teams Actually Use It
1
Get the MCP endpoint
Install the app, open FAC Admin, and copy the site-specific MCP endpoint URL.
2
Add the connector
Register the endpoint in Claude Desktop, ChatGPT connectors, Claude Web, or another MCP client.
3
Authenticate securely
Use the OAuth flow, log in with Frappe credentials, and authorize model access under existing permissions.
4
Execute real ERP work
List customers, create a customer, run a sales report, or query top-selling items from natural language.
Quick Install
Three-Step Setup
# 1. Get the app cd frappe-bench bench get-app https://github.com/buildswithpaul/Frappe_Assistant_Core
# 2. Install on your site bench --site [site-name] install-app frappe_assistant_core # 3. Open FAC Admin and connect your LLM
Example Prompts
Natural Language to ERP Actions
"List all customers in the system" "Create a new customer called Acme Corp with email test@acme.com" "Show me this month's sales report" "What are the top 5 selling items?"
Security & Governance
Enterprise-Ready by Design
AI in ERP systems fails fast when governance is missing. Frappe Assistant Core takes the opposite approach. OAuth 2.0 and OpenID Connect are part of the design, not an afterthought. ERPNext permissions remain the control plane. Audit logging tracks model interactions with ERP data. Role-based access and standardized discovery endpoints make the system safer to deploy and easier to inspect.
That matters for internal teams, implementation partners, and regulated industries. You can give models useful access without treating security as optional.
Security pillars
-
OAuth 2.0 and OIDC support
-
Dynamic Client Registration (RFC 7591)
-
PKCE-enabled flows
-
ERPNext role and permission alignment
-
Audit trail for AI interactions
-
Standardized discovery endpoints
Integration Scenarios
Who This Platform Is For
Business users
Natural language ERP operations through Claude Desktop or other MCP-enabled clients.
Developers
Build custom AI-powered business applications using ERPNext as the operational backend.
System integrators
Deploy reusable LLM-to-ERP solutions across customer environments and industries.
Enterprise teams
Create department-specific AI tools using plugins, policies, and custom business workflows.
Developer and admin enablement
FAC Admin interface: operational control point for configuration and endpoint discovery.
DXT generation: one-click setup path for Claude Desktop integration.
MCP Inspector support: inspect token flow, test tools, and debug request/response patterns.
Plugin-driven extensibility: add custom tools inside Frappe apps or internal plugins.
Admin + Developer Experience
Built for Real-World Deployment, Not Just Demos
The platform includes the operational pieces that serious teams actually need: admin interfaces, testing support, migration guidance, OAuth setup, API references, plugin development paths, and clear onboarding for business users and developers. That is what makes it credible as infrastructure, not just an experiment.
Strategic Value
Why This Matters to Techunison’s Positioning
AI
AI Operating Layer
Positions ERPNext not as a closed application, but as an AI-connected business platform.
op
Operational Leverage
Reduces friction between insight and action by letting models operate directly against governed ERP workflows.
EX
Enterprise Extensibility
Creates a reusable foundation for healthcare, manufacturing, supply chain, finance, and client-specific AI agents.
Build With Techunison
Ready to Connect AI to ERPNext the Right Way?
Frappe Assistant Core shows how AI can move from chat to governed execution. For organizations building the next generation of ERP copilots, departmental agents, and enterprise AI workflows, this is the kind of foundation that matters.