Chart Library MCP Server

MCP server for Chart Library — a visual chart pattern search engine. Find historically similar stock charts from 24M+ patterns across 15K+ symbols and 10 years of data. See what happened next.

What it does

Given a stock ticker and date, Chart Library finds the 10 most similar historical chart patterns using ML embeddings (DINOv2-ViT-B/14) and pgvector similarity search. Returns forward returns (1/3/5/10 day), outcome distributions, and AI summaries.

Tools

Tool	Description
`analyze_pattern`	Recommended — complete analysis in one call: search + forward returns + AI summary
`search_charts`	Find similar historical chart patterns for a ticker + date
`search_batch`	Search multiple symbols at once (up to 20)
`get_follow_through`	Compute 1/3/5/10-day forward returns from matches
`get_pattern_summary`	AI-generated plain-English summary
`get_discover_picks`	Today's most interesting patterns from the daily scanner
`get_status`	Database stats: embeddings, symbols, date range

Install

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "chart-library": {
      "command": "python",
      "args": ["mcp_server.py"],
      "cwd": "/path/to/chart-library-mcp"
    }
  }
}

Claude Code (CLI)

claude mcp add chart-library -- python mcp_server.py

Configuration

Set CHART_LIBRARY_API_KEY for higher rate limits (optional — free tier works without a key):

export CHART_LIBRARY_API_KEY=cl_your_key_here

Get a free key at chartlibrary.io/developers.

Example

User: "What does AAPL's chart from March 20 look like historically?"

Agent uses analyze_pattern("AAPL 2026-03-20"):
→ 10 matches found (JNJ, CVX, MSFT, TJX, RTX...)
→ 5-day forward: avg +0.39%, 6 of 10 went up
→ AI summary: "Based on 10 similar patterns, AAPL shows..."

License

MIT

Chart Library — Pattern Intelligence

Quick Install