MaverickMCP - Personal Stock Analysis MCP Server

MaverickMCP is a personal-use FastMCP server that provides financial data analysis, technical indicators, stock screening, and portfolio tracking tools to any MCP client -- Claude Desktop, Claude Code, Cursor, VS Code, Codex CLI, Antigravity CLI, OpenCode, and others. Built for individual traders and investors, it runs entirely on your own machine with no authentication or billing complexity.

Core tools need no API key: market data comes from yfinance. Two optional extras add more: [backtesting] (VectorBT-powered strategy backtesting) and [research] (LangGraph-based deep research, bring-your-own LLM key).

Skip the setup — hosted version

Self-hosting MaverickMCP means Python, uv, and MCP client config (Redis and a research LLM key are optional). If you just want the analysis, Capital Companion is the hosted product built on the same engine: AI technical analysis, trade-plan review sheets with outcome tracking, and price alerts. 25 free analyses, no credit card.

Self-hosting instructions continue below.

Why MaverickMCP?

Key Benefits:

• No Setup Complexity: make dev gets the server running; no database migrations, no seed scripts, no API key required for core tools.
• Modern Python Tooling: Built with uv for fast dependency management.
• Works With Any MCP Client: Standard MCP server over STDIO or Streamable HTTP -- no client-specific code. See Connect Your MCP Client.
• 37 Core Tools: Market data, technical analysis, screening, portfolio tracking with a risk dashboard, watchlists, and a trade journal.
• Optional Extras: 12 backtesting tools and 3 research tools, each fully opt-in via pip install/uv sync extras.
• Smart Caching: Tiered cache (memory, then Redis or SQLite) with graceful fallback when Redis isn't running.
• Open Source: MIT licensed.

Features

• Stock Data Access: Historical and real-time quotes with intelligent caching (yfinance, no API key required).
• Technical Analysis: RSI, MACD, support/resistance, and a combined full-analysis tool.
• Stock Screening: Maverick bullish, bearish, and supply/demand strategies, computed over the tickers you've already queried.
• Portfolio Tracking: Positions with average cost-basis, live P&L, a risk dashboard, watchlists, and a trade journal.
• Backtesting ([backtesting] extra): VectorBT engine, 12 rule-based strategy templates plus 8 ML strategy classes, optimization, walk-forward analysis, and Monte Carlo simulation.
• Research ([research] extra): LangGraph-based deep research over companies, sectors, and market sentiment, backed by Exa web search and a bring-your-own LLM.
• Multi-Transport Support: STDIO and Streamable HTTP, so any MCP client can connect.

Quick Start

Prerequisites

• Python 3.12+: Core runtime environment
• uv: Modern Python package manager (recommended)
• Redis (optional, for enhanced caching)
• PostgreSQL or SQLite (optional, for data persistence; SQLite is the default)

Installing uv (Recommended)

# macOS/Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

# Alternative: via pip
pip install uv

Installation

Note: v1.0.0 is not yet published to PyPI (registry rollout is in progress). Until it is, use Option 3 (from source). Options 1 and 2 are the intended usage once the package is published.

Option 1: Run without installing (uvx, once published)

# Runs the published maverick-mcp-server package via uvx, invoking its
# maverick-mcp console script
uvx --from maverick-mcp-server maverick-mcp --transport stdio

Option 2: pip install (once published)

pip install "maverick-mcp-server[backtesting,research]"
maverick-mcp --transport stdio

Drop [backtesting,research] for a smaller, core-only install (37 tools, no backtesting/research tools registered).

Option 3: From source with uv (for development)

# Clone the repository
git clone https://github.com/wshobson/maverick-mcp.git
cd maverick-mcp

# Install dependencies and create virtual environment in one command
uv sync --extra dev
# Or, for the full tool surface:
uv sync --extra dev --extra backtesting --extra research

# Copy environment template
cp .env.example .env
# Configure DATABASE_URL / LLM_PROVIDER / EXA_API_KEY as needed (all optional)

Start the Server

make dev          # Streamable HTTP on http://localhost:8003/mcp
make dev-stdio    # STDIO on this terminal

Clients configured for STDIO launch the server themselves -- you do not need make dev running for those.

Connect Your MCP Client

MaverickMCP is a standard MCP server with no client-specific behavior. Any client that speaks the Model Context Protocol can use it: Claude Desktop, Claude Code, GitHub Copilot, Codex CLI, Cursor, OpenCode, Antigravity CLI, and others.

Setup is one decision -- which transport -- followed by pasting the right config shape for your client.

STDIO is the default and the simplest path for one local client. Use Streamable HTTP when several clients should share a single server process.

[!IMPORTANT] The HTTP endpoint has no trailing slash: http://localhost:8003/mcp. /mcp/ returns a 307 redirect, and clients that do not follow redirects on POST will fail to register tools.

Clients not listed still work -- give them the STDIO command or the HTTP endpoint in whatever shape their config expects.

Full reference: docs/runbooks/mcp-clients.md covers every client below in more depth, plus the .mcpb bundle, LAN binding, and per-client troubleshooting. The sections below are the common cases.

Claude Desktop

claude_desktop_config.json launches local STDIO servers only. Using the published package via uvx (no checkout needed):

{
  "mcpServers": {
    "maverick-mcp": {
      "command": "uvx",
      "args": [
        "--from",
        "maverick-mcp-server",
        "maverick-mcp",
        "--transport",
        "stdio"
      ]
    }
  }
}

Running from a local source checkout instead:

{
  "mcpServers": {
    "maverick-mcp": {
      "command": "uv",
      "args": [
        "run",
        "python",
        "-m",
        "maverick.server",
        "--transport",
        "stdio"
      ],
      "cwd": "/path/to/maverick-mcp"
    }
  }
}

Config File Location:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Always fully quit and restart Claude Desktop after making configuration changes.

[!WARNING] Do not paste http://localhost:8003/mcp into Claude Desktop's "custom connector" dialog. Custom connectors are brokered from Anthropic's cloud rather than from your machine, so they cannot reach your localhost. For Claude Desktop, local means STDIO or a .mcpb bundle (make bundle).

Claude Desktop's config file cannot express an HTTP server directly. To use the HTTP transport there, bridge it with mcp-remote (config). It is the main client that still needs that bridge -- every client below speaks Streamable HTTP natively, so do not wrap them in mcp-remote.

[!WARNING] Windows Claude Desktop Users Claude Desktop on Windows currently has a bug where it ignores the "cwd" configuration parameter, which can cause the server to crash with a ModuleNotFoundError when running via uv.

To bypass this, wrap the command in cmd.exe to force the directory change:

"maverick-mcp": {
  "command": "cmd.exe",
  "args": [
    "/c",
    "cd /d C:\\Path\\To\\maverick-mcp && uv run python -m maverick.server --transport stdio"
  ]
}

Claude Code

# Streamable HTTP, against a running `make dev`
claude mcp add --transport http maverick-mcp http://localhost:8003/mcp

# STDIO. The `--` separator is required: without it, `claude mcp add`
# consumes `--transport stdio` as its own flag.
claude mcp add maverick-mcp -- \
  uv run --directory /path/to/maverick-mcp python -m maverick.server --transport stdio

Add --scope user to register the server outside the current project. Verify with claude mcp list.

Cursor

Config Location: ~/.cursor/mcp.json (global) or .cursor/mcp.json (project)

{
  "mcpServers": {
    "maverick-mcp": {
      "url": "http://localhost:8003/mcp"
    }
  }
}

VS Code (GitHub Copilot)

Config Location: .vscode/mcp.json. The key is servers, not mcpServers, and type is required.

{
  "servers": {
    "maverick-mcp": {
      "type": "http",
      "url": "http://localhost:8003/mcp"
    }
  }
}

GitHub Copilot CLI

Config Location: ~/.copilot/mcp-config.json (user) or .mcp.json in the repository. tools filters which tools Copilot exposes; "*" is the default.

{
  "mcpServers": {
    "maverick-mcp": {
      "type": "http",
      "url": "http://localhost:8003/mcp",
      "tools": ["*"]
    }
  }
}

Or: copilot mcp add --transport http maverick-mcp http://localhost:8003/mcp

Codex CLI

Config Location: ~/.codex/config.toml (global) or .codex/config.toml (trusted projects). Shared by the ChatGPT desktop app, Codex CLI, and the IDE extension.

[mcp_servers.maverick-mcp]
url = "http://localhost:8003/mcp"

Or: codex mcp add maverick-mcp --url http://localhost:8003/mcp

OpenCode

Config Location: ~/.config/opencode/opencode.json (global) or opencode.json in the project root. Servers live under mcp, and each entry declares type as remote or local.

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "maverick-mcp": {
      "type": "remote",
      "url": "http://localhost:8003/mcp",
      "enabled": true
    }
  }
}

Antigravity CLI

Google's replacement for Gemini CLI, which stopped serving individual accounts on 2026-06-18.

Config Location: ~/.gemini/config/mcp_config.json (global) or .agents/mcp_config.json (workspace). Remote servers use serverUrl -- the legacy Gemini CLI keys url and httpUrl are not used.

{
  "mcpServers": {
    "maverick-mcp": {
      "serverUrl": "http://localhost:8003/mcp"
    }
  }
}

Or: agy mcp add maverick-mcp http://localhost:8003/mcp

Any Other Client

Nothing client-specific is required: supply either the STDIO command or the HTTP endpoint above. See docs/runbooks/mcp-clients.md for the exact command/args values to paste.

Tools

MaverickMCP registers 37 core tools with a base install. Two optional extras add more. Every tool is read-only (readOnlyHint: true) unless noted otherwise. Full behavior detail lives in ARCHITECTURE.md, docs/features/portfolio.md, docs/features/deep-research.md, and docs/api/backtesting.md.

Market Data (7)

Technical Analysis (4)

Screening (6)

Screens run over the local universe of tickers you've already queried via market-data tools; there is no pre-seeded S&P 500 database. See docs/runbooks/database-setup.md.

Portfolio (20)

All analysis tools auto-detect your portfolio positions when no explicit tickers are supplied. See docs/features/portfolio.md for the cost-basis method and precision rules.

Backtesting (12, [backtesting] extra)

12 rule-based strategy templates plus 8 ML strategy classes. Install with uv sync --extra backtesting or pip install "maverick-mcp-server[backtesting]". Absent the extra, the server still boots and registers zero backtesting_* tools.

Research (3, [research] extra)

Requires EXA_API_KEY (web search) plus a configured BYOK LLM (LLM_PROVIDER/LLM_API_KEY/LLM_MODEL; see Configuration). Install with uv sync --extra research or pip install "maverick-mcp-server[research]". Absent the extra, the server still boots and registers zero research_* tools.

Resources

• portfolio://my-holdings - a passive AI-context snapshot of your default portfolio, automatically available to the assistant.

Prompts

• analyze_stock(ticker) - full technical + screening workflow for one ticker.
• review_portfolio(portfolio_name) - portfolio + risk review workflow.
• run_backtest_workflow(ticker, strategy) - strategy backtesting workflow (registered only with the [backtesting] extra).

Configuration

Configure MaverickMCP via .env file or environment variables. See .env.example for the complete, code-verified list.

Essential Settings:

• DATABASE_URL - PostgreSQL connection or sqlite:///maverick.db for SQLite (default).
• REDIS_HOST - enables Redis caching when set; caching falls back to in-memory/SQLite otherwise.
• LOG_LEVEL - Logging verbosity (default: INFO).

No API key is required to run the core server; stock data comes from yfinance.

Optional (research extra, bring your own key):

• LLM_PROVIDER - anthropic, openai, openrouter, or openai_compatible.
• LLM_API_KEY - API key for the configured LLM_PROVIDER.
• LLM_MODEL - Model name for the configured LLM_PROVIDER.
• LLM_BASE_URL - Base URL override, required when LLM_PROVIDER=openai_compatible.
• LLM_TEMPERATURE - Sampling temperature (default: 0.0).
• EXA_API_KEY - Web search for the research tools (get at exa.ai).

Migrating an older .env (legacy OPENROUTER_API_KEY-style auto-detection, TIINGO_API_KEY, etc.)? See docs/runbooks/migrating-to-v1.md.

Usage Examples

Once connected to Claude Desktop, use natural language:

Technical Analysis

"Show me the RSI and MACD analysis for NVDA"
"Identify support and resistance levels for MSFT"
"Get full technical analysis for AAPL"

Screening

"Run the Maverick bullish screen"
"Show me the top supply/demand breakout setups"

Portfolio

"Add 10 shares of AAPL I bought at $150.50"
"Show me my portfolio with current prices"
"Analyze correlation in my portfolio"  # Auto-detects your positions
"Get my risk dashboard"
"Add AAPL to my watchlist"

Backtesting ([backtesting] extra)

"Run a backtest on AAPL using the momentum strategy for the last 6 months"
"Compare mean reversion vs trend following strategies on SPY"
"Optimize the RSI strategy parameters for TSLA"

Research ([research] extra)

"Research the current state of the AI semiconductor industry"
"Provide comprehensive research on NVDA"
"Analyze market sentiment for the energy sector"

Development

Commands

make dev          # Start server (streamable HTTP transport)
make dev-stdio    # Start server (STDIO transport)
make stop         # Stop services

make test              # Unit tests (fast, default marker filter)
make test-all           # All tests, including integration/slow/external
make test-specific TEST=test_name
make test-watch         # Auto-run tests on file changes

make lint         # ruff check + lint-imports
make format       # ruff format + ruff check --fix
make typecheck     # ty (Astral), same gate as CI
make check          # lint + typecheck
make docs-check      # validate the documentation catalog

# Using uv directly
uv run pytest                 # Unit tests only
uv run pytest --cov=maverick  # With coverage
uv run pytest -m ""           # All tests (requires PostgreSQL/Redis for some)

uv run ruff check .    # Linting
uv run ruff format .   # Formatting
uv run ty check maverick   # Type checking (Astral's ty); same scope as CI

Docker (Optional)

For containerized deployment:

# Copy and configure environment
cp .env.example .env

# Using uv in Docker (recommended for faster builds)
docker build -t maverick-mcp-server .
docker run -p 8003:8000 --env-file .env maverick-mcp-server

# Or start with docker-compose
docker-compose up -d

Note: The Dockerfile uses uv for fast dependency installation. The image ships the [backtesting] and [research] extras by default; drop --extra backtesting --extra research from the uv sync line in the Dockerfile for a smaller, core-only image. There is no HTTP /health endpoint or HEALTHCHECK -- this is an MCP server, not a REST API.

Troubleshooting

Common Issues

Tools missing or disappearing in your client:

• Solution: Use the streamable HTTP endpoint with no trailing slash: http://localhost:8003/mcp
• /mcp/ returns a 307 redirect to /mcp; clients that do not follow redirects on POST fail to register tools
• Also confirm the transport matches: a client configured for STDIO against a running HTTP server (or the reverse) fails silently in most clients
• Per-client troubleshooting: docs/runbooks/mcp-clients.md

Research Tool Timeouts:

• Research tools have adaptive timeouts (120s-600s) based on requested depth
• Deep research may take several minutes depending on complexity
• Monitor progress in server logs with make tail-log

Research Tools Not Available:

• Ensure the research extra is installed: pip install "maverick-mcp-server[research]"
• Ensure LLM_PROVIDER, LLM_API_KEY, and LLM_MODEL are set in .env
• Ensure EXA_API_KEY is set for web search

Backtesting Tools Not Available:

• Ensure the backtesting extra is installed: pip install "maverick-mcp-server[backtesting]"

Empty screening results:

• There is no pre-seeded universe; fetch price history for the tickers you care about first (market_data_get_price_history), then run screening_run_screens. See docs/runbooks/database-setup.md.

# Common development issues
make tail-log          # View server logs
make stop              # Stop services if ports are in use
make clean             # Clean up cache files

# Quick fixes:
# Port 8003 in use → make stop
# Redis connection refused → brew services start redis / unset REDIS_HOST
# Tests failing → make test (unit tests only)

Extending MaverickMCP

Add custom financial analysis tools with simple decorators, following the same pattern used throughout maverick/:

@mcp.tool()
def my_custom_indicator(ticker: str, period: int = 14):
    """Calculate custom technical indicator."""
    # Your analysis logic here
    return {"ticker": ticker, "signal": "buy", "confidence": 0.85}

Getting Help

For issues or questions:

1. Check Documentation: Start with this README, AGENTS.md, and docs/INDEX.md.
2. Search Issues: Look through existing GitHub issues
3. Report Bugs: Create a new issue with details
4. Request Features: Suggest improvements via GitHub issues
5. Contribute: See our Contributing Guide for development setup

Acknowledgments

MaverickMCP builds on these excellent open-source projects:

• FastMCP - MCP framework powering the server
• yfinance - Market data access
• VectorBT - Backtesting engine ([backtesting] extra)
• LangGraph - Research workflow orchestration ([research] extra)
• pandas & NumPy - Data analysis
• The entire Python open-source community

License

MIT License - see LICENSE file for details. Free to use for personal and commercial purposes.

Support

If you find MaverickMCP useful:

• Star the repository
• Report bugs via GitHub issues
• Suggest features
• Improve documentation

Built for traders and investors. Happy Trading!

Read the full build guide: How to Build an MCP Stock Analysis Server

Disclaimer

_{This software is for educational and informational purposes only. It is NOT financial advice.}

_{Investment Risk Warning: Past performance does not guarantee future results. All investments carry risk of loss, including total loss of capital. Technical analysis and screening results are not predictive of future performance. Market data may be delayed, inaccurate, or incomplete.}

_{No Professional Advice: This tool provides data analysis, not investment recommendations. Always consult with a qualified financial advisor before making investment decisions. The developers are not licensed financial advisors or investment professionals. Nothing in this software constitutes professional financial, investment, legal, or tax advice.}

_{Data and Accuracy: Market data provided by third-party sources (Yahoo Finance, and optionally Capital Companion/finviz for market movers). Data may contain errors, delays, or omissions. Technical indicators are mathematical calculations based on historical data. No warranty is made regarding data accuracy or completeness.}

_{Regulatory Compliance: US Users - This software is not registered with the SEC, CFTC, or other regulatory bodies. International Users - Check local financial software regulations before use. Users are responsible for compliance with all applicable laws and regulations. Some features may not be available in certain jurisdictions.}

_{Limitation of Liability: Developers disclaim all liability for investment losses or damages. Use this software at your own risk. No guarantee is made regarding software availability or functionality.}

_{By using MaverickMCP, you acknowledge these risks and agree to use the software for educational purposes only.}