Toolstem is a financial data MCP server designed for AI agents, mainly used for stock analysis, equity research, and agent-driven investment workflows. It is not a simple API passthrough, but instead combines multiple data sources into composite tools, returns flat, agent-friendly JSON, and precomputes derived signals.
Purpose
- Provide AI agents with stock snapshots, company fundamentals, valuation, financial metrics, and company comparison capabilities.
- Suitable for scenarios such as equity research, investment analysis, and automated agent decision-making.
Main Capabilities Provides three read-only, idempotent, safely retryable composite tools:
- get_stock_snapshot: Integrates quotes, company profile, DCF valuation, and ratings into a single response. Derived fields include dcf_signal (DCF upside >10% is UNDERVALUED, <-10% is OVERVALUED, otherwise FAIRLY VALUED), market_cap_readable (e.g. $2.78T), and percentage distance from the 52-week high and low.
- get_company_metrics: Synthesizes profitability, financial health, cash flow, growth, and per-share metrics from 5 financial statement endpoints. period can be annual (default) or quarter. Derived fields include margin_trend (EXPANDING/STABLE/CONTRACTING), health_signal (STRONG/ADEQUATE/WEAK), growth_signal (ACCELERATING/STEADY/DECELERATING), revenue_cagr, net_income_cagr, fcf_cagr, fcf_margin, and fcf_yield.
- compare_companies: Supports side-by-side comparison of 2-5 companies on price, valuation, earnings, financial health, growth, dividends, and analyst ratings, and automatically calculates rankings (lowest P/E, highest margin, strongest balance sheet, best growth, most undervalued, highest rated).
Differences from Common Financial MCP Servers
- Parallel data fetching: Each tool concurrently accesses multiple data sources.
- Derived signals: Directly provides readable recommendations (such as UNDERVALUED, STRONG, ACCELERATING).
- Precomputed math: CAGR, year-over-year growth, margin trends, distance from 52-week high/low, FCF yield, etc. are already included in the response.
- Flat and predictable schema, with no deep nesting.
- Graceful degradation: when a single upstream endpoint fails, the remaining data is still returned, and the failed fields are null.
How to use Using the hosted endpoint is recommended, with no API key and no infrastructure required. Just point the MCP client or agent to the hosted endpoint. initialize and tools/list are free, and each tools/call is paid via x402 at 0.01 USDC (Base mainnet), settled directly from the agent wallet.
- Claude Desktop: configure the hosted endpoint in claude_desktop_config.json, restart, and it can be called, for example "Use Toolstem to get a snapshot of NVDA."
- LangChain.js: use the official @langchain/mcp-adapters to connect to the hosted URL.
- langchain-toolstem wrapper: can automatically handle x402 payments; pass in a funded wallet private key to automatically pay per call. Self-hosting is also supported: run the Node MCP server with your own FMP API key. It supports stdio mode (default, suitable for Claude Desktop, Cursor, etc.) and HTTP mode (local, remote + authentication, local with authentication disabled, remote without authentication, etc.; environment variables need to be set, such as MCP_AUTH_DISABLED=1 to force binding to 127.0.0.1). Self-hosting does not involve x402 and has no per-call fee, but you need to manage FMP_API_KEY and the infrastructure yourself.
Use cases
- Agent-driven stock analysis and investment research.
- Automated workflows that need to quickly obtain multi-source aggregated financial data and directly use derived signals.
- Scenarios integrated with frameworks such as Claude, OpenAI Agents SDK, and LangChain.
Limitations
- All three tools are read-only and do not provide trading or write functionality.
- When self-hosting, you need to obtain your own FMP API key and bear infrastructure responsibility; remote unauthenticated mode has security warnings.
- The hosted endpoint charges per call, 0.01 USDC per call, and requires a Base mainnet wallet.
- More financial MCP servers (SEC filings, insidertrading, institutional holdings) are still under development.
- Other specific features, price changes, platform support, etc. are not provided in the materials.
