Skip to main content
Glama

fmp-mcp

Financial intelligence for your AI agent — powered by live market data.

Give Claude (or any MCP-compatible AI) the ability to analyze stocks, screen markets, compare peers, read earnings calls, and track sector rotations — all grounded in real financial data from Financial Modeling Prep.

This isn't a raw API wrapper. Each tool is purpose-built for a specific analytical task, composing multiple data sources into structured, analysis-ready outputs designed for AI consumption.

The source server registers 22 MCP tools (including imported tool registrations). The estimate-revision tools read the hosted estimates API; the estimates extra and PostgreSQL are needed for local estimate storage, not hosted MCP reads.

What your AI can do

Market Intelligence

  • get_market_context — One-call market snapshot: indices, sectors, gainers/losers, economic events

  • get_sector_overview — Sector and industry performance with P/E valuations

  • get_news — Stock-specific or broad market news

  • get_events_calendar — Earnings, dividends, splits, and IPO calendars

  • get_economic_data — Economic indicators and high-impact event tracking

Fundamental Analysis

  • fmp_fetch — Direct access to 60+ financial data endpoints (income statements, balance sheets, cash flows, key metrics, and more)

  • fmp_market_cap_check — Compare current market cap against the latest annual filing value

  • compare_peers — Side-by-side peer comparison across 12 financial ratios

  • get_earnings_transcript — Parsed earnings calls with speaker attribution and Q&A sections

  • get_stock_fundamentals — Enriched stock lookup with financials, valuation, quality, and technicals

  • get_price_performance_windows — Compact benchmark-relative return windows

Stock Screening & Discovery

  • screen_stocks — Screen by sector, market cap, beta, dividend, volume, and more

  • get_institutional_ownership — Institutional holder analytics and ownership trends

  • get_insider_trades — Insider transaction flow and statistics

  • get_etf_holdings — ETF/fund holdings, sector and country allocation

Technical Analysis

  • get_technical_analysis — Composite signals from 7 indicators (SMA, EMA, RSI, MACD, Bollinger, ADX, Williams %R) with buy/sell scoring

Analyst Sentiment

  • get_estimate_revisions — Historical EPS/revenue estimate revision trends

  • screen_estimate_revisions — Screen for estimate momentum across a universe of stocks

Data Discovery

  • fmp_list_endpoints — Browse all 60+ available data endpoints by category

  • fmp_describe — Get parameter documentation for any endpoint

  • fmp_search — Search for companies by name or ticker

  • fmp_profile — Company profile with sector, industry, and key stats

Related MCP server: Financial Modeling Prep MCP Server

Install

pip install fmp-mcp

Optional local estimate storage (requires PostgreSQL; hosted MCP reads do not):

pip install "fmp-mcp[estimates]"

Configuration

Set your API key:

export FMP_API_KEY="your_key"

Optional settings:

  • FMP_CACHE_DIR — Custom cache directory (default: ~/.cache/fmp-mcp/)

  • FMP_CACHE_MAXSIZE — Max in-memory cache entries (default: 200)

Application integration

The installed fmp-mcp distribution owns the fmp import. It does not search for a Risk checkout, load a checkout .env, or import application policy. The source package lives at Risk's fmp-mcp/. Install it from the Risk root with the single editable list, .venv/bin/python -m pip install --no-deps -r requirements-local.txt, which also installs brokerage-connect. Dev tools come from requirements-dev.txt.

Applications must explicitly configure the client once at startup before any network fetch. An unconfigured client refuses dispatch and names the required configure_client(...) call. Applications can bind these callbacks:

  • fmp.client.configure_client(budget_guard=..., event_observer=...). The guard accepts fn, args, kwargs plus provider/operation/budget metadata. The observer receives (event, endpoint, **details), where event is success (duration_ms, status), rate_limit, error (error), plan_limit (error), or cache_hit (params). Explicit configure_client() selects the public default: fmp.spend writes one machine-wide API spend-ledger row per network call and a zero-call cache_hit row per cache-answered request (hank-model-authority), refuses a call whose spend ceiling is already met with SpendCeilingExceeded, and rate limits and errors go to native Python logging. A host that passes its own budget_guard owns metering and nothing in fmp.spend runs. The public fmp-mcp / python -m fmp.server entrypoint selects the default policy before serving tools.

  • fmp.tools.peers.configure_peer_tools(metric_resolver=..., peer_discovery=..., spot_fx_rate=...). Metric resolvers return the existing structural peers/source/to_dict() result; discovery returns ticker strings. Expected policy failures raise PeerSelectionError. Standalone discovery uses FMP stock_peers; metric-context selection without application policy asks for explicit peers. Standalone non-USD conversion reads the latest daily close of FMP's <currency>USD pair. The optional FX callback returns a currency-to-USD rate.

Risk process entrypoints call fmp_runtime.configure() after environment hydration to bind all five callbacks; logging, billing, peer selection, and exchange-mapping policy remain application code. bootstrap_env.bootstrap() only hydrates environment and never imports FMP or application policy. scripts/run_fmp_server.py owns Risk credential hydration and policy setup before invoking the installed fmp.server.run() (which preserves the caller's policy). Risk-only wrappers live in utils/fmp_compat.py and utils/fmp_fx.py. The former local estimates collector was deleted in fee0d9403; see estimate operations for the hosted API and production collector.

The source retains the Hank-only transcript_kpi_fetcher.py and manifest_source_dispatcher.py; public distribution sync excludes them. Their application consumers require the separately installed named industry and value-semantics libraries. Sync copies source bytes without vendoring or rewriting imports.

Run

fmp-mcp

Or register the installed package with Claude Code from any directory:

claude mcp add fmp-mcp --scope user \
  -- python3 -m fmp.server

You can also use a generic MCP config:

{
  "mcpServers": {
    "fmp-mcp": {
      "type": "stdio",
      "command": "uvx",
      "args": ["fmp-mcp"],
      "env": { "FMP_API_KEY": "your_key" }
    }
  }
}

How it's different

Raw API wrapper

fmp-mcp

Approach

Expose every endpoint 1:1

Purpose-built analytical tools

Output

Raw JSON, dozens of fields

Structured, summarized, analysis-ready

Composition

One API call per tool

Multiple sources stitched together

AI-optimized

Generic descriptions

Tool descriptions and schemas designed for LLM tool selection

Caching

None

Per-endpoint disk caching with configurable refresh strategies

Requirements

Available Tools

21 tools
compare_peersA

Compare a stock against its peers on key financial ratios.

Fetches the peer group for a stock and builds a side-by-side comparison of financial ratios including valuation, profitability, margins, and leverage. By default, auto-discovery is operating-peer oriented. For dividend payout, buyback, and shareholder-return questions, pass peer_context="capital_allocation" or metric="dividend_payout" so the peer set favors mature dividend-paying capital-allocation comps over no-dividend growth operating peers.

Inputs: symbol (required), limit, peers, format, +3 more. Returns: object

fmp_describeA

Get detailed documentation for an FMP endpoint.

Discovery tool to understand what parameters an endpoint accepts. Use this before calling fmp_fetch to know required vs optional params.

Inputs: endpoint (required). Returns: object

fmp_fetchA

Fetch data from any registered FMP endpoint.

This is the primary data fetching tool. Use fmp_list_endpoints to discover available endpoints, and fmp_describe to see required parameters.

Inputs: endpoint (required), type, year, limit, +9 more. Returns: object

fmp_list_endpointsA

List available FMP data endpoints.

Discovery tool to see what data is available. Use this first to understand what endpoints exist before fetching data.

Inputs: category. Returns: object

fmp_market_cap_checkA

Compare current profile market cap against latest annual key_metrics market cap.

Use this when fmp_profile.mktCap and fmp_fetch(endpoint="key_metrics") marketCap values appear to disagree. The profile value is a current snapshot; the latest annual key_metrics value is aligned to that row's period-end date.

Inputs: symbol (required). Returns: object

fmp_profileA

Get detailed company profile information as a current FMP /profile snapshot.

Returns comprehensive company data including sector, industry, description, CEO, employees, website, and key financial metrics. The profile mktCap field is price times shares as of the API call time, or up to 1 week stale when served from cache. FMP does not include a snapshot timestamp on this endpoint. For historical period-aligned market cap, use fmp_fetch(endpoint="key_metrics", period="annual") and read the date field on each row.

Inputs: symbol, ticker. Returns: object

get_earnings_transcriptA

Parse and navigate an earnings call transcript.

Splits the raw transcript into structured sections: prepared remarks and Q&A with per-speaker segments and grouped Q&A exchanges.

IMPORTANT — default mode is summary (metadata only). This protects your context window. Full text is returned only when you explicitly request format="full".

Recommended workflow:

  1. Call with default format="summary" to see the speaker list, word counts, and exchange count (costs ~1 KB of context)

  2. Identify the section or speaker you need

  3. Call again with format="full" and specific filters (section, filter_role, filter_speaker) to read only that content

  4. Each text field is capped at max_words (default 3000). Set max_words=None to remove the cap (use with caution). Note: If format="full" is used WITHOUT any filters (no section, filter_role, or filter_speaker), a bounded preview is returned instead (first 3 segments per section, 500 words each, 2 exchanges). Add at least one filter to get full content with max_words control.

Inputs: year (required), symbol (required), quarter (required), format, +5 more. Returns: object

get_economic_dataA

Get economic indicators or upcoming economic events.

Fetches macroeconomic data from FRED via FMP. Use indicator mode for time series data (GDP, CPI, unemployment, etc.) or calendar mode for upcoming/recent economic releases.

Inputs: mode, limit, format, output, +5 more. Returns: object

get_estimate_revisionsA

Get estimate revision history for one ticker and fiscal period.

Reads snapshots through the hosted estimates API backed by the monthly collection job. If fiscal_date is omitted, defaults to the nearest available upcoming fiscal period for the ticker.

Inputs: ticker (required), period, fiscal_date. Returns: object

get_etf_holdingsC

Get ETF/fund composition across holdings, sector/country weights, and metadata.

Inputs: symbol (required), limit, format, output, +2 more. Returns: object

get_events_calendarA

Fetch upcoming corporate events: earnings, dividends, splits, or IPOs.

Can show market-wide calendars or filter to specific symbols. Useful for tracking upcoming catalysts and corporate actions. Date ranges wider than 90 days are split into FMP-compatible windows internally. Note: For portfolio-aware auto-fill, use get_portfolio_events_calendar on portfolio-reads-mcp.

Inputs: limit, format, symbols, to_date, +2 more. Returns: object

get_insider_tradesB

Get insider buying/selling activity and summary statistics for a stock.

Fetches recent insider transactions and aggregate insider trading stats.

Inputs: symbol (required), limit, format, use_cache. Returns: object

get_institutional_ownershipA

Get institutional ownership analytics for a stock.

Combines institutional holder-level data with aggregate ownership summary metrics to show which institutions are accumulating or reducing positions.

Inputs: symbol (required), year, limit, format, +2 more. Returns: object

get_market_contextA

Get a one-call market snapshot across indices, sectors, movers, and events.

Useful for quick "what's happening in the market?" checks. Fetches major index moves, sector heatmap, top gainers/losers, most active names, and upcoming high-impact economic events.

Inputs: format, include, use_cache. Returns: object

get_newsA

Fetch news articles for stocks or the broad market.

Three modes: stock-specific news, general market news, and company press releases. Requires symbols for stock and press modes. Note: For portfolio-aware auto-fill, use get_portfolio_news on portfolio-reads-mcp.

Inputs: mode, limit, format, quality, +3 more. Returns: object

get_price_performance_windowsA

Get compact benchmark-relative price-performance windows.

Use this for historical-coincidence evidence when a workflow needs 3-5 outperformance/underperformance periods. It fetches adjusted historical prices for the stock and benchmark, aligns common trading dates, computes rolling returns, and returns only the largest non-overlapping windows. This avoids raw historical-price CSV parsing for ordinary stock-move analysis.

Inputs: symbol (required), to_date, direction, from_date, +5 more. Returns: object

get_sector_overviewA

Get sector or industry performance and valuation overview.

Combines daily performance snapshot and P/E ratio snapshot into a unified heatmap view. Supports both sector-level and industry-level granularity.

Inputs: date, level, format, sector, +2 more. Returns: object

get_technical_analysisA

Get composite technical analysis for a single stock or ETF.

Fetches multiple technical indicators and provides trend direction, momentum signals, volatility analysis, and support/resistance levels with an overall buy/sell signal.

Inputs: symbol (required), format, last_n, output, +4 more. Returns: object

screen_estimate_revisionsA

Screen a ticker universe for material estimate momentum.

Compares latest estimates against a lookback snapshot (days ago) and returns per-ticker deltas/direction. By default, trivial revisions are filtered out unless either EPS changed by at least 1% of baseline consensus or revenue changed by at least 0.5% of baseline consensus. If tickers is omitted, screens the full stored universe.

Inputs: days, period, tickers, direction, +3 more. Returns: object

screen_stocksA

Screen stocks by fundamental criteria (sector, market cap, beta, dividend, etc.).

Searches the full universe of stocks and ETFs using financial filters. Combine multiple criteria to narrow results.

Inputs: limit, format, is_etf, sector, +14 more. Returns: object

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.9/5.0

    Scored across 21 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose, covering specific financial data domains (e.g., profile, transcripts, insider trades, economics, screening). Even overlapping areas like fmp_profile and fmp_fetch are differentiated by description.

    Naming Consistency3/5

    Tool names are inconsistent: some use the 'fmp_' prefix (fmp_fetch, fmp_describe) while others use 'get_' (get_news, get_earnings_transcript) or 'screen_' (screen_stocks). This mixed convention may confuse an agent.

    Tool Count5/5

    With 21 tools covering a broad range of financial data (company, market, technical, events, ownership, etc.), the count is well-scoped for a comprehensive financial data MCP server.

    Completeness5/5

    The tool surface covers major financial analysis areas: profile, peers, market context, price performance, technical, sector, news, events, earnings transcripts, insider trades, institutional ownership, ETF holdings, economic data, and screening. The generic fmp_fetch endpoint fills any remaining gaps.

    Maintenance

    ActivityMaintained
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      C
      quality
      B
      maintenance
      Enables AI assistants to access and analyze financial data including stock information, company fundamentals, and market insights through the Financial Modeling Prep API.
      267
      360 npm
      149
      TypeScript
      Apache 2.0
    • A
      license
      Not graded
      quality
      D
      maintenance
      Provides AI assistants access to comprehensive financial data including real-time stock quotes, company fundamentals, financial statements, market analysis, economic indicators, and 250+ financial tools across 24 categories from Financial Modeling Prep API.
      Apache 2.0
    • F
      license
      A
      quality
      F
      maintenance
      Provides financial data from Financial Modeling Prep for AI-assisted investment research, including company profiles, financial statements, and analyst ratings. It features high-level workflow tools for market analysis and atomic tools for deep dives into valuation and institutional ownership.
      21
      -
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables Claude to access real-time stock data, financial statements, cryptocurrency information, and market data via the Financial Modeling Prep API.
      4
      MIT