Purpose
Chart Library MCP is the MCP server implementation of Chart Library, positioned as "market memory for AI." It connects visual stock chart pattern search capabilities to MCP-compatible clients or agents, enabling agents to find historical price movements similar to a specified stock, date, or screenshot, and to see what happened next in those historical analogies.
Main Capabilities
The server provides five read-only tools. Market status requests require no prior search or queue handle; they can directly read the latest built/published trading session status by stock ticker (with optional date), outputting historical analogy intervals, sample sizes, session dates, and limitation notes. The date format is YYYY-MM-DD; when the date is omitted, the latest built/published session is used rather than a real-time quote. Published research requires first searching for an ID, then reading its evidence; the returned content includes document sections, reading parameters, and guidance to continue reading via content.next_read. Search summaries serve only as discovery metadata, not as source reads; keyword search ranks document relevance, not market similarity or evidence strength.
Usage
The primary remote endpoint is https://chartlibrary.io/mcp , using Streamable HTTP, with no authentication required. For command-based clients, the corresponding Python package can be installed or upgraded, and an equivalent connection configured in the MCP client. The server requires no API key or server-side Python package; the optional CHART_LIBRARY_API_KEY, when set, is forwarded as a Bearer token, while an empty key sends no Authorization header. CHART_LIBRARY_API_URL can be used to select your own compatible endpoint, but should only point to a trusted server. Running requires Python 3.10+, and the current version uses the MCP 1.x FastMCP interface, pinned to mcp>=1.28.1,<2.0.0. You can either call MCP tools or use equivalent HTTP calls directly; historical status can append date=YYYY-MM-DD, and in daily REST requests this parameter is named session.
Applicable Scenarios
Suitable for enabling AI agents to conduct well-documented market pattern research: reading the historical analogy range, sample size, and limitations for a stock's latest completed trading session, or reading evidence documents by ID around published research. It can be used to build development integrations or larger-scale research based on historical pattern memory, but the public menu does not expose private fund operations or all experimental state memory methods.
Limitations
The data is not real-time quotes; when the date is omitted, the latest built/published session is used. Missing or weak evidence remains missing or weak. A changed published version returns an error and explains how to restart from the current overview; clients do not automatically retry. Research access is free but still subject to service limits and underlying data terms; HTTP 429 and retry guidance should be followed, and unlimited throughput should not be assumed. Historical frequency does not constitute buy or sell advice, and raw historical percentiles are not automatically calibrated predictions. research_quality only audits the methods and population specified in its receipt, and its coverage percentage should not be repurposed for all market states or all research. Daily research has its own selection rules and settled note counts.
