Modernizing U2 UniData: From DICTionaies/UniBasic code to a Live CRM in Minutes using UOFastCoder Skills & Agents + MCP.

Kurt @ RokiPark — Sun, 19 Apr 2026 19:58:50 +0000

UOFastMCP + UOFastCoder let AI agents read your live U2 UniData schema and generate a complete Python Flask (WTForms) based CRM app.

If you're running a business on U2 UniData, your data is already there — CLIENTS, CONTACTS, ORDERS, INVENTORY. What's missing is a modern interface your team can actually use.

UOFastMCP and UOFastCoder are two open-source tools that let an AI agent read your live UniData schema and generate a complete React CRM — no manual BASIC coding, no bespoke REST adapters.

Here's what the finished result looks like:

Real data. Real app. Let's build it.

What you need

A running UniData instance with the DEMO account
Python 3.11+ and Node.js 20+
Claude Code (or any MCP-capable AI client) If using VSCode with Claude Extension, then please read README within UOFastCoder and UOFastMCP ---

How it works

Your AI Agent (Claude Code)
        │
        ▼
  UOFastMCP Server          ← connects to live UniData via uopy
        │
        ▼
  UniData DEMO Account      ← CLIENTS · CONTACTS · ORDERS · INVENTORY
        │
        ▼
  UOFastCoder Skills        ← generates TypeScript types + React components
        │
        ▼
  Your React CRM App        ← running at localhost:5173

Step 1 — Install UOFastMCP

git clone https://github.com/rokipark/uofastmcp.git
cd uofastmcp
pip install -r requirements.txt

Follow instruction in (https://github.com/rokipark/uofastmcp)

Illustration below

Start the server:

uofast-mcp

Step 2 — Connect Claude Code to UOFastMCP

this adds connection string to your ~/.claude/settings.json:

claude mcp add --transport sse UOFastMCP http://localhost:8000/sse \
  --header "Authorization: Basic <your-base64-token>"

Claude Code will now have access to your live UniData files as MCP tools — query_file, read_record, get_dict_items, and more.

Step 3 — Let UOFastCoder generate the app

Verify the connection /uo-setup

Confirms the MCP server is reachable and credentials are correct. If anything is wrong it gives exact steps to fix it.

Document your database (one-time setup)
/uo-document --all
Next, Open Claude Code and type this single prompt:


Using UOFastMCP and UOFastCoder, introspect the CLIENTS,
CONTACTS, and ORDERS files in the demo connection, then
generate a complete React CRM app with:
  - A dashboard with summary cards and recent records
  - A paginated, searchable Clients list with view and edit
  - An Orders view with line items
  - An Express API gateway wiring everything together

UOFastCoder reads your actual DICT definitions — field names, attribute numbers, conversion codes like MD2 for currency and D4/ for dates — and generates strongly-typed TypeScript interfaces and React components that match your real data structure.

Step 4 — Run it

cd crm-app
python app.py

That's it. Your CRM is live at http://localhost:<Port>, reading directly from UniData.

What gets generated

The finished app includes everything you saw in the screenshots above:

Dashboard — Clients and Orders counts, quick-action cards, recent records at a glance
Clients CRUD — searchable paginated table, view detail, inline edit, New Client form
Orders — order list with client lookup and line item detail
Inventory — parts and pricing from your INVENTORY file
Express gateway — typed REST routes, multi-value field expansion, date conversion

All driven by your actual UniData DEMO data — Bank and Trust, Travelbooks, United Hospital, and the rest of the records you already have.

Go further

Once you have the DEMO working, pointing at your own account is one line change in connections.json. UOFastCoder's skills also cover:

Write-back forms — save edits back to UniData via write_record
** ( not part of this demo ) SB+ screen extraction** — convert legacy Process screens to React
Docker deployment — containerize the full stack for production

Both tools are open source. Star the repos and check out the docs at rokipark.ai.

🔗 rokipark/uofastmcp
🔗 rokipark/uofastcoder

U2 UniData with Low-code AI: UOFast + VectorShift RAG Integration

Kurt @ RokiPark — Sat, 05 Jul 2025 04:30:26 +0000

Introduction

A new way to develop a pipeline that connects legacy UniData (U2) systems with modern AI using a Python library called UOFast, enabling seamless data extraction and feeding that into a Retrieval-Augmented Generation (RAG) app powered by VectorShift. If you're looking to modernize legacy MultiValue databases and make them AI-friendly with Low-code RAG, read on!

🔧The Problem

Many businesses still rely on Rocket's U2 (UniData/UniVerse) systems — battle-tested but difficult to integrate with modern AI workflows. The challenge? Bridging that gap with minimal disruption and enabling value extraction from data stored in a non-relational, MultiValue world.

💡 The Solution: UOFast + VectorShift

I created a project that connects U2 UniData to a RAG app using:

UOFast: A lightweight Python library that connects to UniData via Python uopy & extracts data into Python-native formats.

VectorShift: A low-code RAG platform that lets you build AI chatbots backed by your own data (PDFs, structured files, etc.).

🔗 Architecture Overview

UOFast pulls data from UniData, flattens MultiValue structures, and prepares documents.
These documents are ingested into VectorShift, where they're indexed and used in a RAG pipeline.
The end result? A natural language interface to query legacy system data using GPT-powered search!

Steps

- Startup RokiPark/UOFast to connect to your local U2 Unidata.

(Please see Github link) for more info.

- Connect to the UOFast API from VectorShift to extract information

VectorShift Documentation link

- Create a Chatbot interface in VectorShift

Tip - You can use a template from the existing Vectorshift templates to create this RAG workflow. I used a CSV template and replaced the CSV data extraction with UOFast API extraction.

Once completed, a chatbot link is created in Vectorshift which can be embedded in your web application. Here is a Demo which uses the Unidata/Demo CLIENT table to answer questions

Demo link

🤖 Why RAG?

Retrieval-Augmented Generation brings your private data into GPT-style answers without retraining. By plugging in UniData's structured output, I could enable:

Natural language search for records

Chatbot Q&A on sales history, order data, etc.

Summarization of legacy reports

🔜 What’s Next

Add real-time sync for live querying

Support Program generation and Auto-API conversion (auto-generating RESTful APIs from legacy logic)

Build a lightweight admin panel with FastAPI + NiceGUI for non-technical users

📢 Feedback Welcome!

This is a work in progress, and I’d love feedback, especially if you work with UniData or build enterprise RAG solutions. PRs, feature requests, and use-case suggestions are welcome on GitHub or email - tech@rokipark.ai

🔗 Links:

RokiPark UOFast

VectorShift Documentation

UniData by Rocket Software

Let’s bring legacy systems into the AI era — one table at a time. 🧠💾

DEV Community: Kurt @ RokiPark