TabulaRAG
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@TabulaRAGShow me sales data for Q1 2024 with citations"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Features
CSV/TSV ingestion — upload tabular files with automatic header detection, delimiter inference, and column type recognition (dates, money, measurements)
Semantic search — each row is embedded via
sentence-transformers/all-MiniLM-L6-v2and indexed in Qdrant for natural-language retrieval with cell-level citationsStructured queries — filter and aggregate data through the API with SQL over PostgreSQL JSON columns
Multi-tenant workspaces — enterprises with invite codes, roles (owner / admin / querier), and switchable active workspace
Folders & access control — public, protected, and private folders with user-group-based permissions
MCP server — exposes Streamable HTTP + SSE endpoints so AI assistants can use your tables as a retrieval tool
Auth — Google OAuth and email/password with verification codes and password reset (via Brevo SMTP)
Background indexing — threaded worker pool for non-blocking embedding and Qdrant upserts with progress tracking
Related MCP server: mcp-excel
Architecture
Layer | Technology |
Frontend | React 19, TypeScript 5.9, Vite 7, React Router 7 |
Backend | Python 3.11, FastAPI, SQLAlchemy 2.x, Uvicorn |
Database | PostgreSQL 16 |
Vector store | Qdrant v1.13.4 (FastEmbed, 384-dim dense vectors) |
Auth | JWT, Google OAuth, bcrypt |
MCP | fastapi-mcp (Streamable HTTP + SSE) |
Web server | Nginx 1.27 (production frontend) |
CI | GitHub Actions (pytest, ESLint, Docker builds) |
MCP integration
TabulaRAG exposes endpoints for AI assistant integration:
Type | URL |
OpenAPI |
|
MCP (Streamable HTTP) |
|
Authentication: use a personal MCP token (generated from the app's MCP section) in Authorization: Bearer <token>, or the server API_KEY for automation. Tokens are scoped per user and workspace.
If your MCP client runs outside the browser (e.g. Docker, desktop app), replace
localhostwith your machine's IP (ipconfigon Windows,ifconfigon Mac/Linux).
Local vs. Deployed
Feature | Local (Docker Compose) | Deployed |
Database | PostgreSQL in Docker | External PostgreSQL |
Vector store | Qdrant in Docker | External Qdrant |
Prerequisites (Local)
Tool | Version | Notes |
Latest | ||
v2+ | Bundled with Docker Desktop |
Docker Compose v2 is required (
docker compose, notdocker-compose). It ships with Docker Desktop on Mac and Windows. Linux users may need to install it separately.
You do not need to install Python or Node locally — everything runs inside containers.
Quick start
cp .env.example .env # create config (edit values as needed)
./scripts/dev-up.sh # build and start all servicesOnce running:
Service | URL |
Frontend | |
Backend API | |
PostgreSQL | localhost:5433 |
Qdrant |
Health checks:
curl http://localhost:8000/health
curl http://localhost:8000/health/depsStop / logs:
./scripts/dev-down.sh
./scripts/dev-logs.sh # all services
./scripts/dev-logs.sh backend # single serviceEnvironment variables
Copy .env.example to .env. The key variables:
Variable | Required | Description |
| Yes | Database name (default: |
| Yes | Database user (default: |
| Yes | Database password |
| Yes | PostgreSQL connection string (set in docker-compose) |
| Yes | Qdrant endpoint (set in docker-compose) |
| Yes | Secret for signing JWT tokens |
| Yes | Backend URL used in email links (default: |
| Yes | Frontend URL used in email links (default: |
| No | Optional static key for script/API access |
| No | Google OAuth client ID |
| No | Google OAuth client secret |
| No | Brevo SMTP relay for verification/reset emails |
| No | SMTP port (default: |
| No | SMTP login |
| No | SMTP key |
| No | Sender email address |
SMTP is optional for local development. If
SMTP_HOSTis empty, verification and reset codes are written to the backend logs instead of being emailed. For any real deployment, configure SMTP so users receive emails.
The remaining variables in .env.example control embedding model tuning, Qdrant HNSW parameters, batch sizes, and indexing concurrency. The defaults work for local development.
License
This server cannot be deployed
Maintenance
Related MCP Connectors
- OleanderOAuthdev.oleander
The all-in-one data stack for agents. Upload files, run SQL, evolve tables, and render charts.
Ask data questions in natural language. Get SQL, insights, and charts from your databases.
Query your warehouse or a CSV with Claude/ChatGPT over MCP, governed by table-level ACL + audit.
Ask questions in plain language, get answers from your business database. No SQL required.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceEnables users to ask questions about their spreadsheets in plain English and receive instant answers using DuckDB-powered analytics, supporting multiple file formats like Excel, CSV, and Parquet.2MIT
- AlicenseNot gradedqualityDmaintenanceEnables querying Excel and CSV files using SQL via natural language, allowing AI assistants to analyze data without manual SQL writing.1MIT
- AlicenseNot gradedqualityDmaintenanceTransforms AI assistants into data analysts by enabling CSV and Google Sheets import, SQL querying, and instant insights through natural language.MIT
- FlicenseNot gradedqualityBmaintenanceEnables AI assistants to present interactive spreadsheets within conversations, allowing users to view, edit, import, and export Excel/CSV data directly through the chat interface.-