AI

Frappe AI

AI in the Awesome Bar

Frappe AI puts an assistant where your users already are: the Awesome Bar. It augments global search with natural-language understanding of your ERP data, and connects to an MCP server over OAuth2 so answers come from your live site rather than a model's memory.

Language
Python · JavaScript
License
MIT
Runtime
Frappe app
Auth
OAuth2 client
Backend
MCP Server
View on GitHub
QUICKSTART
Install
cd ~/frappe-benchbench get-app https://github.com/vyogotech/frappe_aibench --site your-site install-app frappe_aibench restart
Configure
# Create the OAuth client and copy the credentials it printsbench --site your-site execute frappe_ai.setup.create_oauth_client # Then in the Desk UI: /app/mcp-server-settings#   Enabled                 -> checked#   MCP Server URL          -> http://localhost:8080#   Frappe Base URL         -> http://localhost:8000#   OAuth Client ID/Secret  -> paste from above# Save, then hit Test Connection.

What Frappe AI gives you

Awesome Bar, upgraded

The search box every Frappe user already uses becomes a natural-language interface — find documents, ask questions, get answers without learning a new surface.

Context-aware results

Results are grounded in your actual data model and the current user's workspace, so relevance comes from your site rather than a generic embedding.

Permission-preserving

The app overrides the whitelisted global-search method rather than bypassing it, so Frappe's role and document permissions still decide what a user can see.

Wired to MCP

Configured through MCP Server Settings with an OAuth2 client, with a built-in Test Connection so you know the link works before users do.

Under the hood

Frappe AI overrides the global search whitelisted method to return AI-augmented results while leaving Frappe's permission and security model untouched — the assistant can never surface a document the user could not already open.

Read the source

Frequently asked questions

Does Frappe AI need the MCP Server?

Yes. Frappe AI is the in-Desk experience; the MCP Server is the backend that talks to your data and your model. You configure the server URL and an OAuth2 client under MCP Server Settings.

Can users see data they don’t have access to?

No. The app hooks into Frappe's whitelisted global-search method rather than going around it, so role permissions, document-level permissions and user permissions all still apply.

How long does setup take?

About five minutes: install the app, run one bench command to create the OAuth client, paste the credentials into MCP Server Settings, and press Test Connection.

Works well with

MCP Server

AI bridge for ERPNext

A Model Context Protocol server that lets Claude, Cursor and any LLM query ERPNext safely.

Explore

Krama

Project planning for Frappe

Hierarchical task planning and interactive Gantt charts built on Frappe's own datatable.

Explore

SaaS Platform

Multi-tenancy on a single site

A Frappe app that lets one site serve many tenants, isolated by row instead of by database.

Explore

Running Frappe AI in production?

We build and maintain these tools, and we help teams adopt them. Tell us what you are trying to ship and we will tell you honestly whether this is the right piece.

Talk to the team