使用说明

MCP Chart Generator

MCP (Model Context Protocol) server that exposes the Trading Lab Chart Generator functionality, allowing Claude to generate professional trading charts on demand.

Overview

This MCP server provides programmatic access to the trading-lab chart generation system, enabling:

On-demand chart generation during Claude conversations
28+ technical studies (RSI, MACD, Bollinger Bands, etc.)
ML predictions from Chronos, TOTO, Prophet, Burg models
Signal overlays with buy/sell markers
Multiple themes (Dark Casual, Dark Professional, Light, etc.)
Flexible timeframes (Daily, Weekly, Monthly, Yearly)

Architecture

Claude Code
    ↓ (MCP Protocol)
MCP Chart Generator Server
    ↓ (Python Import)
trading-lab/src/services/chart_generation/
    ↓
PNG Chart Image

Available Tools

1. `render_chart`

Generate a trading chart PNG image.

Required Parameters:

symbol_code - Stock symbol (e.g., "AAPL", "SPY")
file_name - Output filename without extension

Optional Parameters:

exchange_code - Exchange (default: "US")
symbol_type - Type (default: "Common Stock")
study - Study name (default: "ChartOnly")
theme - Chart theme (default: "Dark Casual")
chart_type - Candlestick/Line/HLC (default: "Candlestick")
period - Daily/Weekly/Monthly/Yearly (default: "Daily")
history_length - Days of history (default: 365)
warmup_period - Extra bars for indicators (default: 200)
zoom_level - Number of bars to display (default: 250)
width - Image width in pixels (default: 1200)
height - Image height (default: auto)
prediction_mode - off/view_only/full (default: "view_only")
bullish_signal - Bullish signal for overlays
bearish_signal - Bearish signal for overlays
marker_source - actual/predicted (default: "actual")
eod_historical_offset - Days back from today (0-30)

Example:

mcp__chart-generator__render_chart(
    symbol_code="AAPL",
    file_name="aapl_rsi",
    study="RSIStudy",
    zoom_level=120,
    theme="Dark Casual"
)

2. `list_studies`

List all available chart studies.

Returns: Categorized list of 28+ studies

3. `list_signals`

List all available trading signals.

Returns: Bullish and bearish signal names

4. `list_themes`

List all available chart themes.

Returns: Available theme names

5. `validate_config`

Validate chart configuration before rendering.

Parameters:

symbol_code - Stock symbol
study - Study name (optional)
theme - Theme name (optional)
bullish_signal - Bullish signal (optional)
bearish_signal - Bearish signal (optional)

Returns: Validation status and errors if any

Installation

Prerequisites

Python 3.11+
trading-lab project available at expected path
MARKET_DATABASE_PATH or MARKETDATA_DB_PATH environment variable set

Setup

Create virtual environment:

cd /Volumes/Work/development/projects/git/mcp-chart-generator
python3 -m venv venv

Install dependencies:

venv/bin/pip install -r requirements.txt

Configure in .mcp.json (in trading-lab project):

{
  "mcpServers": {
    "chart-generator": {
      "type": "stdio",
      "command": "/Volumes/Work/development/projects/git/mcp-chart-generator/venv/bin/python3",
      "args": [
        "/Volumes/Work/development/projects/git/mcp-chart-generator/chart_generator_mcp_server.py"
      ],
      "env": {
        "MARKET_DATABASE_PATH": "/Volumes/Work/marketdata/marketdata.db"
      }
    }
  }
}

Test the server:
```
cd /path/to/trading-lab
claude mcp list
```
You should see chart-generator in the list with ✓ Connected status.

Usage Examples

Example 1: Simple Chart

User: "Generate an AAPL chart"

Claude uses:
mcp__chart-generator__render_chart(
    symbol_code="AAPL",
    file_name="aapl_chart"
)

Output: /tmp/aapl_chart.png

Example 2: Chart with RSI Study

User: "Show me SPY with RSI indicator, last 90 days"

Claude uses:
mcp__chart-generator__render_chart(
    symbol_code="SPY",
    file_name="spy_rsi_90d",
    study="RSIStudy",
    zoom_level=90
)

Output: /tmp/spy_rsi_90d.png

Example 3: Chart with Signal Overlays

User: "MSFT chart with buy/sell signals using ADM momentum"

Claude uses:
mcp__chart-generator__render_chart(
    symbol_code="MSFT",
    file_name="msft_signals",
    bullish_signal="adm_momentum_low",
    bearish_signal="adm_momentum_high",
    zoom_level=180
)

Output: /tmp/msft_signals.png with buy/sell markers

Example 4: Discover Available Options

User: "What chart studies are available?"

Claude uses:
mcp__chart-generator__list_studies()

Returns: List of all 28+ studies categorized by type

Viewing Generated Charts

Important: Claude Code currently does not support inline image display (iTerm2 image protocol).

After generating a chart, you have two options to view it:

Option 1: Exit and use imgcat (iTerm2)

# Exit Claude Code session (Ctrl+D or /exit)
imgcat /tmp/your_chart.png

Option 2: Open in default viewer

# From within Claude Code or after exiting
open /tmp/your_chart.png

This will open the chart in Preview (macOS default image viewer).

Why this happens

Claude Code intercepts terminal output and doesn't currently pass through iTerm2's inline image escape sequences. This is tracked in GitHub issue #6389.

Workaround: The MCP server generates charts successfully; viewing them just requires opening them outside the Claude Code session.

Technical Details

Dependencies

The MCP server itself only requires:

mcp>=2,<3 - MCP protocol implementation

The actual chart generation is handled by importing from the trading-lab project, which has its own dependencies (pandas, playwright, etc.).

Output Location

Generated charts are saved to /tmp/ directory with the filename specified in the file_name parameter.

Performance

Chart generation takes 1-2 seconds per chart
The server reuses resources for efficiency
Can generate hundreds of charts in batch

Integration Points

Market Data:

Requires MARKET_DATABASE_PATH environment variable
Uses trading-lab's marketdata.database.Database

Studies:

Auto-discovers from trading-lab/src/studies/
Supports all 28+ studies

Signals:

Uses trading-lab/src/signals/bullish_signals/
Uses trading-lab/src/signals/bearish_signals/

Troubleshooting

Server won't connect

Check virtual environment:

ls -la /Volumes/Work/development/projects/git/mcp-chart-generator/venv/bin/python3

Verify dependencies:

/Volumes/Work/development/projects/git/mcp-chart-generator/venv/bin/pip list | grep mcp

Check trading-lab path: Ensure /Volumes/Work/development/projects/git/trading-lab/src exists
Verify environment variable:
```
echo $MARKET_DATABASE_PATH
```

Chart generation fails

Check database:
```
ls -la $MARKET_DATABASE_PATH
```

Test trading-lab directly:

cd /Volumes/Work/development/projects/git/trading-lab
source venv/bin/activate
python tests/test_chart_generation_poc.py

Validate configuration: Use validate_config tool before render_chart

Development

Running Tests

Test the MCP server independently:

cd /Volumes/Work/development/projects/git/mcp-chart-generator
venv/bin/python3 chart_generator_mcp_server.py

This will start the server in stdio mode (waiting for MCP protocol messages).

Adding New Tools

Edit chart_generator_mcp_server.py:

Add tool definition in list_tools()
Add handler in call_tool()
Implement the async function

Debugging

Enable debug logging in the trading-lab:

import logging
logging.basicConfig(level=logging.DEBUG)

Related Projects

trading-lab - Main trading analysis platform
mcp-marketdata-db - Market data MCP server
mcp-strategy-research-db - Strategy backtest MCP server
mcp-finance - Yahoo Finance MCP server

License

Same as trading-lab project.

Author

Created as part of the Trading Lab ecosystem for programmatic chart generation via MCP.