Hybrid Bayesian + GPT-2 Synthetic Data Generator
DataMerase MCP Server
Generate synthetic data by asking Claude Desktop directly — no code required. The MCP server wraps the DataMerase API in six tools that Claude can call on your behalf.
What it is

MCP (Model Context Protocol) lets Claude Desktop talk to external tools during a conversation. The DataMerase MCP server is a small local script that bridges Claude to the https://datamerase.io API: point it at a CSV on your machine, describe which columns are the target / continuous / categorical / text / ignore, and ask Claude to generate synthetic rows — it uploads the file, starts the job, polls it to completion, and saves the result, all inside the chat.

Tools it exposes
check_server_health
Checks whether the DataMerase API is up and how many jobs are currently running — useful before starting a new one.
upload_dataset(file_path)
Uploads a local CSV/Excel file and returns a job_id plus auto-detected column types and data-quality warnings.
generate_synthetic_data(job_id, ...)
Starts a generation job for an uploaded dataset, given a role for every column (target / continuous / categorical / text / ignore).
check_job_status(job_id)
Polls a job's status, progress log, and fidelity metrics once it finishes.
download_results(job_id, save_path)
Downloads the finished synthetic_data.csv to a local path.
generate_and_wait(file_path, save_path, ...)
The whole pipeline in one call: upload → generate → poll to completion → download. What most prompts end up using.
Installation
  1. Install dependencies — needs Python 3.10+.
    pip install mcp requests
  2. Download the server script below and save it somewhere permanent, e.g. ~/datamerase_mcp.py.
    ⬇ Download datamerase_mcp.py
  3. Get an API key — contact admin@datamerase.io if you don't have one yet.
  4. Add it to your Claude Desktop config at ~/Library/Application Support/Claude/claude_desktop_config.json (macOS), %APPDATA%\Claude\claude_desktop_config.json (Windows), or ~/.config/Claude/claude_desktop_config.json (Linux). Merge this into the mcpServers object — don't overwrite existing entries:
    claude_desktop_config.json
    { "mcpServers": { "datamerase": { "command": "python3", "args": ["/absolute/path/to/datamerase_mcp.py"], "env": { "DATAMERASE_API_URL": "https://datamerase.io", "DATAMERASE_API_KEY": "your-datamerase-api-key-here" } } } }
  5. Restart Claude Desktop. DataMerase's tools should now show up in the tool picker.
Example prompts

Once it's connected, just ask — Claude figures out which tools to call.

"Generate 1000 synthetic rows from my housing.csv file with SalePrice as target and Gr Liv Area as continuous"
"Upload my sales_data.csv, treat Region and Category as categorical, ignore the CustomerID column, and generate 500 synthetic rows — save it to ~/Desktop/synthetic_sales.csv"
"Check if the DataMerase server has capacity for a new job right now"