MaverickMCP

What it is

MaverickMCP is a FastMCP server you run on your own machine and point an MCP client at — Claude Desktop, Claude Code, Cursor, Codex CLI, VS Code. A base install registers 37 tools: 7 for market data, 4 for technical analysis, 6 for screening, and 20 for a local portfolio with cost-basis P&L, a risk dashboard, watchlists and a trade journal. Two opt-in extras add 12 VectorBT backtesting tools and 3 LangGraph research tools that want your own LLM key and a search backend — Exa by default, or a self-hosted SearXNG instance with no key.

Against calling yfinance yourself, it buys three things: the indicator maths already written (RSI, MACD, stochastic, ATR, ADX, moving averages), a SQLite file that remembers positions, watchlists and journal entries between sessions, and the MCP plumbing so an agent reaches all of it without a wrapper. The data underneath is the same Yahoo data, with the same failure mode.

v1.1.0 shipped on 5 September 2026 — the second tagged release — and commits were still landing on the default branch on 13 September. One maintainer, 668 stars, not archived.

Pricing

Free, MIT-licensed, no account. It is not on PyPI: the name maverick-mcp-server belongs to an unrelated dormant project and a PEP 541 transfer is pending, so you install from a git tag with uvx, from the ghcr.io image, or from a checkout. Do not guess the shorter name: maverick-mcp on PyPI is a different publisher's package with nothing to do with this project.

Data & coverage

All market data reaches the server through yfinance against Yahoo Finance — unauthenticated and unofficial, which is exactly why no key is needed. The screening and portfolio tools read the local database and touch no network at all. The market overview is hard-coded to six US indices and eleven US sector ETFs, though it is configurable; price history, quotes and fundamentals follow any ticker Yahoo knows, including non-US listings. Market movers fall through three tiers: the Capital Companion API if you hold a key, a finviz screener scrape, then a batch quote over a short fixed list of liquid names.

Integrations

STDIO or Streamable HTTP on localhost:8003/mcp — no trailing slash, because /mcp/ answers 307 and clients that do not follow redirects on POST register nothing and say nothing. SQLite by default, PostgreSQL and Redis optional, Python 3.12+. There is no REST API and no health endpoint.

Limitations

• The screens ship without a universe. They compute over tickers already in the local database, and the Tiingo bulk loader that used to seed an S&P 500 list was removed at v1.0.0.
• One upstream, no fallback. When Yahoo throttles by IP or moves an endpoint, every tool that needs a fresh price stops; the circuit breaker and 5-requests-per-second default limiter make that survivable, not avoidable.
• Capital Companion — the hosted product the README offers as the no-setup alternative — closes on 19 September 2026, which also retires the first movers tier.
• Positions are typed in by hand. No broker connection, no import, no tax reporting, no options.
• Nothing reaches you. The risk alerts are computed when a tool is called; there is no scheduler and no notification channel.
• Some outputs are directional labels on named tickers: an outlook banded from "strongly bullish" to "strongly bearish", and a position line reading "Consider averaging down" or "Consider taking partial profits". That is threshold arithmetic over indicators, not a model, and the project's own disclaimer calls the whole thing educational.

Alternatives

Vendor-run MCP servers from Alpha Vantage, EODHD or Financial Modeling Prep trade the no-key convenience for licensed data and someone to complain to. The same author's mcp-trader is a smaller server over the same ideas. If you only want the numbers in a notebook, call yfinance directly and skip the server.

Compare it against the rest of the MCP servers and AI agents.