Plenty of tools will hand a model a P/E ratio. Here the method behind each number is open-source, and awkward details — fiscal-year alignment, currency conversion for a ticker listed on the wrong exchange — are handled explicitly rather than silently. The FMP key is the real gate: hosted or local, the data comes from your subscription.
The MCP server shipped with the Finance Toolkit, a Python library of financial calculations. It exposes those 500+ methods to any MCP client, so the assistant fetches live financial data and runs ratios and models whose implementations you can read in the repository.
- 500+ financial methods callable from the conversation, over live data, with the calculations open-source rather than a black box
- Financial statements normalised to calendar periods, so companies with different fiscal years are comparable
- Currency conversion for tickers listed outside their home exchange, keeping ratios like free cash flow yield and P/E correct; disabled with `convert_currency=False`
- Trailing and growth variants on most functions, combinable — set `trailing=4` on quarterly data for TTM figures, `growth=True` for period-over-period, with an optional lag
- A local SQLite cache shared with the library, which retrieves only what is missing when you widen a period or add a ticker, and never caches values the toolkit computes itself
A Financial Modeling Prep API key. Two ways in. Hosted: connect to `https://financetoolkit.jeroenbouma.com/mcp` — the first connection opens an OAuth consent page asking for your FMP key, and nothing is installed locally. Local: run `uvx --from "financetoolkit[mcp]" financetoolkit-mcp-setup`, which finds your client's config and writes the entry including the key, or configure `financetoolkit-mcp` by hand with `FINANCIAL_MODELING_PREP_API_KEY` set. `FRED_API_KEY` is optional and unlocks a handful of US-only economic indicators. An MCPB bundle exists for one-click Claude Desktop install. Version 2.2.0.