erpnext.2nth.ai — Platform Online

ERPNext is the OS
for your AI agents.

Spin up a fully wired ERPNext instance in minutes. Connect AI agents via MCP. Build systems where one person does the work of 2n.

human + ERP data + AI agents = 2n output
See Member Tiers Watch the Deploy
<10m
Instance Ready
$0
When Idle
MCP
Agent Protocol
2n
Leverage
The Foundation

AI agents are only as good
as their data source.

Without structured, transactional ground truth, agents hallucinate. ERPNext gives them a real system of record to read, write, and reason over — with permissions, audit trails, and state that persists.

Agents without ERP
  • ✗ No ground truth — agents invent data they can't find
  • ✗ No state — every session starts from scratch
  • ✗ No permissions — agents either see everything or nothing
  • ✗ No audit trail — impossible to know what an agent actually did
  • ✗ Brittle integrations — scrapers and webhooks that break on every update
ERPNext as Agentic OS
  • ✓ Structured reality — every customer, order, invoice is a queryable DocType
  • ✓ Persistent state — agents read and write the same data humans do
  • ✓ Role-scoped MCP tools — agents get exactly the permissions they need
  • ✓ Full audit log — every agent action is a traceable transaction
  • ✓ Stable REST + MCP API — the same interface, forever

How an ERP instance becomes an OS

Every module in ERPNext is a structured subsystem. Agents connect to the ones they need, read live state, and write back — just like processes on an OS.

Accounts
Financial State
GL entries, invoices, payments — real-time cash position for any agent that needs it.
Stock
Physical Inventory
Bin-level quantities, reorder points, projected demand — the ground truth for procurement agents.
CRM
Customer Graph
Contacts, interactions, pipeline — structured context for sales and support agents.
Projects
Work in Progress
Tasks, timesheets, milestones — live operational state for delivery and billing agents.
2nth Member Tiers

Pick your starting point.

Every tier gives you a real ERPNext instance and access to the agent layer. The difference is how deep you go — and how many agents you're running.

Explorer
Sandbox
Learn the agentic ERP pattern in a safe environment with no production consequences.
  • ▸ Shared ERPNext sandbox with sample data
  • ▸ 3 pre-built agents to explore (inventory, invoicing, data quality)
  • ▸ MCP connection guide for Claude / Cursor
  • ▸ Access to 2nth skills catalog
  • − Dedicated instance (shared)
  • − Custom agent workflows
Request Access
Most Popular
Builder
Your Instance
A dedicated ERPNext instance, spun up for your business or project — your data, your agents.
  • ▸ Dedicated ERPNext instance on GCP africa-south1
  • ▸ Industry data pack pre-loaded (SA localised)
  • ▸ MCP endpoint + API keys for your agents
  • ▸ Connect Claude, Cursor, n8n, or your own agents
  • ▸ Scales to zero when idle (~R900/mo base)
  • ▸ Cloudflare Zero Trust access control
Request a Demo
Enterprise
Multi-Instance
Run multiple instances for clients, departments, or staging environments — with production SLAs.
  • ▸ Everything in Builder, for multiple instances
  • ▸ Agent gateway for cross-instance orchestration
  • ▸ Custom MCP tools and DocType extensions
  • ▸ HA Cloud SQL + automated backups + SLA
  • ▸ White-label for client delivery
  • ▸ Direct Slack/email support from 2nth team
Talk to Us
The 2n idea

One person. Exponential systems.

2nth is built on the belief that AI agents should multiply what one person can do — not replace them. An ERPNext instance is the shared memory that makes this real.

You configure, approve, and direct. Agents monitor inventory, draft invoices, check data quality, and report back. The ERP is the substrate they all share.

// What 2^n looks like in practice
human (1)
manages → ERP instance
directs → 3 agents
reviews → agent drafts
approves → final actions
// agents running 24/7 in background
inventory-agent: monitoring
invoicing-agent: drafting
data-quality-agent: scanning
// effective output of 1 person
= 2n operations/day
Architecture

Built for zero-ops, infinite scale.

Every component scales independently. Idle demos cost nothing. Production instances get HA and automated backups. Click any component to learn more.

Edge Layer
Compute Layer — Google Cloud Run
VPC Connector
Data Layer — Private VPC
Secret Manager
Credentials + API keys
Artifact Registry
Docker images
IAM + Terraform
Least privilege IaC

What it costs

~R900
/month idle demo
Cloud SQL + Redis
Compute: R0 (scale to zero)
~R1,200
/month active demo
Light usage
Pay-per-request compute
Custom
/month production
HA + backups + SLA
Scales with your usage
Live Simulation

Watch it happen.

This is what it looks like when 2nth provisions a complete ERP instance for a client. Pick an action and watch the terminal.

2nth-erp ~ deploy
$
Ready 0%
Agent Layer

Agents that work on real data.

Every 2nth instance ships with AI agents connected via MCP (Model Context Protocol). They read your DocTypes, take action, and log everything — then wait for your approval before anything is submitted.

▣

Inventory Agent

inventory-agent

Monitors stock levels across warehouses. Flags items below reorder points, drafts purchase orders, and alerts on projected stockouts before they happen.

▸ Monitors Bin, Stock Entry, Purchase Order
▸ Creates draft POs when stock is low
▸ Runs on schedule or on-demand via MCP
read: Bin read: Item write: Purchase Order
□

Invoicing Agent

invoicing-agent

Drafts sales invoices from delivery notes and timesheets. Checks for missing details, calculates VAT correctly, and queues invoices for review — never sends without approval.

▸ Monitors Delivery Note, Timesheet
▸ Drafts invoices with correct SA VAT (15%)
▸ Human-in-the-loop — always draft, never submit
read: Delivery Note read: Timesheet write: Sales Invoice
◇

Data Quality Agent

data-quality-agent

Scans for data issues — duplicate customers, missing tax IDs, orphaned records, inconsistent naming. Reports findings weekly and suggests fixes you can apply with one click.

▸ Scans Customer, Supplier, Item, Account
▸ Detects duplicates, missing fields, orphaned links
▸ Weekly digest with actionable fix suggestions
read: Customer read: Supplier read: Item
MCP — Model Context Protocol

MCP is an open standard for connecting AI models to external tools and data sources. Every 2nth ERP instance exposes MCP tools that agents use to read DocTypes, create records, run reports, and trigger workflows — through a structured, auditable interface.

No scraping. No brittle API wrappers. Agents operate through the same permission system as human users, with role-based access and full audit trails.

// Agent permission model
agent: "inventory-agent"
role: "Inventory Manager"
can_read: [Bin, Item, Warehouse]
can_write: [Purchase Order (draft)]
can_submit: false
audit_log: true
Build your own agents

The 3 bundled agents are a starting point. Your ERPNext instance exposes the same MCP interface to any agent you build — in Claude, Cursor, n8n, or your own code.

Claude Code — connect directly, ask questions, build automations
n8n / Make — trigger agents from DocType events, webhooks
Custom Python/TS — use the Frappe REST API + MCP server directly
Cloudflare Workers — edge agents that rate-limit, cache schema, proxy
Every agent you add compounds your output.
1 person + n agents = 2n
Get Started

Get your instance. Connect your agents.

Tell us your industry and tier. We'll spin up your ERPNext instance with sample data and send you the MCP endpoint — ready for Claude, Cursor, or your own agents from day one.

1

Choose your tier + industry

Explorer (sandbox), Builder (your own instance), or Enterprise. Pick an industry data pack: professional services, manufacturing, retail, or NGO.

2

Instance spins up

A full ERPNext environment on GCP africa-south1 — under 10 minutes, pre-loaded with SA-localised data. You get the URL and your MCP endpoint.

3

Connect your agents

Point Claude, Cursor, n8n, or your own code at the MCP server. The 3 bundled agents are already running — add more to multiply your output.

Learn more about 2nth.ai → View on GitHub →