Rowmind MCP Server & Dashboard
Rowmind Hero

⚠️ CLONE REQUIRED FOR TESTING

This is a single-user, private-instance project. The public demo is limited because exposing CSV files would let anyone delete or corrupt data.

To test properly: Clone the repo → Run locally or as a private Space → Connect your MCP client.

This design ensures your data stays yours.

What is this?

Rowmind is a small "memory layer" for your AI chat client. You talk to Claude, ChatGPT, Cursor... and they can save stuff for you. Tasks, expenses, random notes, whatever you want.

The twist? Everything is stored in plain CSV files. No fancy database. No cloud sync. Just files you can open with Excel or any text editor.

Why CSV?

  • You own your data — it's right there, on your disk
  • Easy to backup — just copy the folder
  • Easy to debug — open the file and see what's inside
  • Works offline — no internet needed for storage
  • Hackable — add columns, edit rows, do whatever

How it works

 
┌─────────────────┐          ┌─────────────────┐          ┌─────────────────┐
│   Your Chat     │   MCP    │    Rowmind      │  R/W     │   CSV Files     │
│   Client        │ <------> │    Server       │ <------> │   (your data)   │
│ (Claude, etc)   │ Protocol │   (this app)    │          │                 │
└─────────────────┘          └─────────────────┘          └─────────────────┘

[TODO: Add nicer diagram image here]

The AI uses MCP protocol to talk to this server. When you say "remind me to call mom tomorrow", the AI:

  1. Calls list_tables() to see what tables exist
  2. Picks the right one (probably tasks)
  3. Calls append_row() to save it

The dynamic part (this is the cool thing)

You can upload any CSV and the AI will automatically know how to use it.

How? There's a special tables.csv that stores metadata about each table:

┌─────────────────────────────────────────────────────────────────────────┐
│                           tables.csv                                     │
├─────────────┬─────────────────┬─────────────────────────────────────────┤
│  table_name │   description   │            example_sentence             │
├─────────────┼─────────────────┼─────────────────────────────────────────┤
│  expenses   │ Daily spending  │ "I spent 15 dollars on lunch"           │
│  movies     │ Films to watch  │ "Add Inception to my movie list"        │
│  mood       │ How I feel      │ "Today I'm feeling tired"               │
│  recipes    │ Cooking ideas   │ "Save this pasta recipe"                │
│  ...        │ ...             │ ...                                     │
└─────────────┴─────────────────┴─────────────────────────────────────────┘
                                    ↓
                    AI reads this and understands:
                    "oh, when user talks about money → expenses.csv"
                    "when user mentions films → movies.csv"

So when you say "I spent 20 bucks on coffee", the AI:

  1. Checks tables.csv to see what tables exist
  2. Matches your message to the right table (expenses, because of the money context)
  3. Appends a row to expenses.csv

🎯 You can track anything.

Workouts, books, habits, dreams, whatever. Just upload a CSV with your columns and add a description. The AI figures out the rest.

Features

What How
Tasks & Reminders Schedule one-time or recurring reminders. Get push notifications via ntfy
Custom Tables Create any table you want — expenses, mood tracker, whatever
SQL Queries Query your CSVs with DuckDB. Yes, real SQL on CSV files
Table Metadata Add descriptions so the AI knows what each table is for

The notification thing

 
┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│  tasks.csv  │ --> │ Background  │ --> │    ntfy     │ --> 📱 Phone
│  (pending)  │     │   Worker    │     │   (push)    │
└─────────────┘     └─────────────┘     └─────────────┘

A background job checks every minute for due tasks and sends push notifications through ntfy. No Firebase, no Apple stuff, just HTTP requests.

Demo Videos

See Rowmind in action:

📊 Dynamic CSV Tables

Upload any CSV and the AI automatically learns how to use it. Query with SQL, append rows, update fields — all through natural conversation.

⏰ Tasks & Push Notifications

Schedule one-time or recurring reminders. A background worker monitors due tasks and sends push notifications via ntfy — no Firebase needed.

Privacy note

This demo on Hugging Face is public — anyone can see the data.

For real use, duplicate this Space and make it private. Or run locally. The whole point is that YOUR data stays with YOU.

Twitter

📢 Original announcement on Twitter

Built for the Gradio MCP Hackathon 🚀