pychartai

Natural-language data analysis and chart generation. Use it as a Python library, a CLI, or an MCP server for Claude/Copilot/Cursor.

Quick Start

pip install pychartai

import pychartai as pai

# Set DEEPSEEK_API_KEY in .env, then:
pai.init(provider="deepseek", model="deepseek-chat")

df = pai.read_csv("data.csv")
print(df.chat("What is revenue by region?"))
path = df.chat("Bar chart of revenue by region")  # interactive HTML
print(df.profile().summary)                        # auto-EDA, no LLM

Three Ways to Use It

1. MCP Server (for Claude Desktop / Copilot / Cursor)

make mcp-server
# or: pychartai-mcp

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "pychartai": {
      "command": "python",
      "args": ["-m", "pychartai_mcp.server"],
      "env": {"DEEPSEEK_API_KEY": "sk-..."}
    }
  }
}

Then in Claude: "Load sales.csv and show revenue by region as a chart"

Available MCP tools: analyze_data, create_chart, profile_data

2. Python Library

import pychartai as pai

pai.init(provider="deepseek", model="deepseek-chat")

sdf = pai.read_csv("sales.csv")
print(sdf.chat("Top 3 products by revenue?"))
path = sdf.chat("Bar chart of revenue by region")

# Streaming, memory, profiling
for event in sdf.chat_stream("Summarize this"):
    if event.type == "token": print(event.value, end="")
sdf.enable_memory()
sdf.chat("Follow-up question")

3. CLI

pychartai data.csv "Revenue by region?"
pychartai data.csv --profile          # auto-EDA, no LLM
pychartai data.csv "Bar chart" --backend plotly

Features

Installation

pip install pychartai

# Optional extras:
pip install pychartai[viz-plotly]   # interactive Plotly charts
pip install pychartai[db-postgres]  # PostgreSQL connector

Set your API key in .env:

DEEPSEEK_API_KEY=sk-...
OPENAI_API_KEY=sk-...

Configuration

models.yml (auto-discovered)

default_model: deepseek-v4-flash
models:
  deepseek-v4-flash:
    provider: deepseek
    model: deepseek-chat
    api_key_env: DEEPSEEK_API_KEY
settings:
  chart_backend: plotly

pai.init()                              # auto-discover + default_model
pai.init(model="deepseek-v4-flash")     # by name
pai.init(provider="deepseek", model="deepseek-chat")  # inline

Makefile Reference

Setup:     make install    make venv    make prepare-data
Offline:   make test       make demo    make gallery
With LLM:  make demo-advanced [PROVIDER=deepseek]
MCP:       make mcp-server

License

MIT