Building Financial Data Agents with MCP

← Back to shibui.finance

MCP (Model Context Protocol) connects AI models directly to external data sources. Shibui Finance is a free MCP server with 64 years of US stock market data already loaded: daily prices, quarterly financials, technical indicators, and SEC filings for 10,000+ US equities. These guides show you how to wire that database into automated pipelines, scheduled screens, and multi-step analysis workflows. Each includes connection config, scheduling commands, and working examples.

Quick start

Two commands to go from nothing to automated screening results in a file.

1. Register the MCP server:

claude mcp add shibui-finance --transport streamable-http https://mcp.shibui.finance/mcp

2. Run a screen and save the output:

claude -p "Connect to Shibui Finance. Find the 10 largest US stocks by market cap that closed below their 200-day SMA yesterday. Return as a JSON array with keys: symbol, name, market_cap, price, sma_200." \
  --allowedTools 'mcp__shibui-finance__*' > screen-results.json

That is a complete automated screen. The -p flag runs Claude non-interactively with the prompt, --allowedTools grants access to the Shibui MCP tools, and stdout is redirected to a file. Wrap it in a cron job or a /schedule command and it runs daily.

For Claude on the web, ChatGPT, or other MCP clients, see the connection options in each workflow guide below.

Why MCP instead of a REST API

A REST API approach to the quick-start screen above would require calls to at least two endpoints (one for market cap and valuations, one for technical indicators), pagination handling for 10,000+ symbols, response parsing, and join logic to merge results by ticker. That is 50+ lines of Python before you see a single result. The MCP approach is one English sentence through a single connection.

The maintenance difference compounds over time. API scripts break when providers change schemas, rename fields, or adjust rate limits. MCP queries adapt because the model reads the database schema at connection time. If a column name changes, the model sees the new name. You do not update a parser.

Shibui's database has 31 million price rows, fundamentals going back 35 years, and 56 technical indicators computed daily. All of that is pre-joined and indexed. The model can filter across tables, compute aggregates, and return sorted results in one query. The individual workflow guides cover the details: output format control, prompt reliability for unattended runs, error handling, and multi-step chaining.

Workflow guides

Related guides

Looking for interactive screening guides instead of automated workflows?

• AI Stock Screener - screen the full US market in plain English
• How to Screen Stocks with AI - step-by-step screening walkthrough
• VCP Screener - Minervini's volatility contraction pattern
• CAN SLIM Screener - O'Neil's 7 criteria for growth stocks
• Free Financial Data API for AI - why MCP is replacing REST APIs
• All guides and comparisons

Frequently asked questions

What is an MCP server for financial data?

MCP (Model Context Protocol) connects AI models directly to external data sources. Shibui Finance is a free MCP server with a pre-loaded database of 10,000+ US equities: 64 years of daily prices, quarterly financials, 56 technical indicators, and SEC filings. Any MCP-compatible client can query it in plain English without API keys, without writing code, and without uploading spreadsheets. The model reads the schema at connection time and writes the query itself.

Is Shibui Finance free?

Yes. The MCP server at mcp.shibui.finance/mcp is free with no API key and no subscription; fair-use limits apply. You pay only for your AI client (Claude, ChatGPT, etc.), and free tiers work. The data is updated daily and covers NYSE and NASDAQ equities and ETFs.

What clients can connect to Shibui via MCP?

Any MCP-compatible client: Claude (web, desktop, and Code), ChatGPT, Codex, and custom agents. Connect via streamable HTTP transport to https://mcp.shibui.finance/mcp. No authentication required. The server exposes query (run SQL against the database), get_database_schema (load the schema), get_query_patterns (load query templates), and workflow loaders for fundamentals, technicals, screening, comparisons, earnings, backtesting, filings, and insider transactions.

What data is available for automated screening?

Daily OHLCV prices for 10,000+ NYSE and NASDAQ securities back to 1962, quarterly and annual financials back to 1993, daily valuations (market cap, P/B, enterprise value, trailing P/E, EV/EBITDA), 56 pre-computed technical indicators (RSI, MACD, Bollinger Bands, SMA, EMA), SEC insider transactions from Forms 3/4/5, and analyst estimates including forward PEG. Full inventory on the data sources page.