Reviewed overview

MCP サーバー概要

what is MaverickMCP?

MaverickMCP is a self-hosted FastMCP server for personal stock analysis, technical indicators, stock screening, portfolio tracking and optional backtesting/research workflows. It exposes 37 core read-only tools (market data, TA, screening, portfolio) and optional backtesting and research toolsets as MCP endpoints you can connect to Claude Desktop, Cursor, Windsurf, or the Claude CLI. Core market data uses yfinance and the project is MIT-licensed.

how to use MaverickMCP?

Quick start: clone the repo or install the package (uv/uvx or pip). Typical dev command: make dev to start the server (streamable HTTP on http://localhost:8003/mcp/ by default) or run with STDIO for direct Claude Desktop integration.
Install extras: add [backtesting] and/or [research] via uv sync --extra or pip install "maverick-mcp-server[backtesting,research]" to enable backtesting (VectorBT) and research (LangGraph + BYOK LLM + EXA web search).
Configure with a .env or environment variables (DATABASE_URL, REDIS_HOST optional, LLM_PROVIDER/LLM_API_KEY/EXA_API_KEY only required for research extras).
Connect clients: recommended STDIO for Claude Desktop (or use mcp-remote bridge for clients expecting STDIO while server is HTTP). Docker image and development commands are provided.

key features

Market data (OHLCV, quotes, fundamentals) with smart caching (memory → Redis/SQLite fallback) via yfinance.
Technical indicators: RSI, MACD, support/resistance, full combined analysis.
Screening: bullish/bearish/supply-demand screens computed over tickers you queried.
Portfolio management: positions, cost-basis, live P&L, watchlists, trade journal, risk dashboard, position sizing tools.
Optional backtesting extra: VectorBT engine, strategy templates, ML strategy classes, optimization, walk-forward, Monte Carlo.
Optional research extra: LangGraph-driven web-search-backed company/sector research and sentiment (requires EXA and a BYOK LLM).
Multi-transport: streamable HTTP and STDIO; native Claude Desktop integration.
Developer-friendly tooling: uv, pytest, lint/typecheck, Docker, extendable via simple MCP tool decorators.

use cases

Quick technical analysis and signal generation for individual tickers inside Claude Desktop.
Build and track a personal portfolio, log trades, and run risk-adjusted sizing and correlation analysis.
Run backtests and parameter optimizations (with backtesting extra) to evaluate strategies before paper/live trading.
Perform deep web-search-backed research and sentiment analysis on companies/sectors (with research extra + BYOK LLM).
Embed custom finance tools or extend the toolset for automated workflows and reports.

FAQ from the project

Q: Do I need API keys to run the core server? A: No. Core tools use yfinance and require no API key. Research/backtesting extras may need external keys (EXA_API_KEY, LLM_API_KEY) and optional Redis/Postgres for persistence.
Q: Why are some tools missing in my client? A: Ensure the HTTP endpoint includes a trailing slash (/mcp/). Also confirm the relevant extras are installed—backtesting_* and research_* tools register only when their extras are installed.
Q: How do I connect to Claude Desktop? A: STDIO is recommended. Example JSON config is provided in the README. For HTTP servers, use mcp-remote as a bridge.
Q: Research/backtesting tools time out or fail—what now? A: Research tasks can take minutes; monitor logs. Ensure EXA_API_KEY and LLM settings are configured for research. Install the backtesting extra to enable backtesting tools.
Q: Is this financial advice? A: No. The README includes a clear disclaimer: educational/informational only. Not investment advice; use at your own risk.
Q: Can I run this in Docker / enterprise? A: Yes—Dockerfile and docker-compose examples are included. The image can include extras by default or be slimmed down.
Q: License and contribution? A: MIT licensed. Contribution guidance and issue templates are provided on GitHub.