cvi-mcp — CVI global equity intelligence for AI assistants

An MCP server that makes the CVI Platform a native tool for Claude Desktop, Claude Code, Cursor and any MCP-capable assistant. Ask in plain language — "the 20 most undervalued stocks worldwide", "Continental's margin of safety", "what did CVI's model say about NEM on July 15?" — and the assistant calls CVI's research engine directly: 25 markets, ~2,900 companies, trailing-twelve-month valuations, hierarchical model predictions and a point-in-time Time Machine.

Every tool response carries provenance — the immutable, checksummed research run and as-of timestamp behind the numbers — so the assistant can cite exactly where a figure came from.

Install

pip install cvi-mcp

Connect

Claude Desktop — add to claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "cvi": {
      "command": "cvi-mcp",
      "env": { "CVI_API_KEY": "cvi_live_…" }
    }
  }
}

Claude Code:

claude mcp add cvi -e CVI_API_KEY=cvi_live_… -- cvi-mcp

Generate a key from your CVI account. Scopes map to plan tiers — see pricing.

Tools

Tool	What it answers	Scope
cvi_screen	Most undervalued stocks worldwide (filter by market/sector/signal)	read
cvi_markets	Which 25 markets are covered	read
cvi_market_rankings	Ranked securities within one market	read
cvi_company	One company's IDs + scores + valuation + margin of safety	read
cvi_time_machine	The research snapshot recorded for a past date	research
cvi_history	How CVI's view of a company evolved across runs	research
cvi_predictions	Hierarchical model picks with confidence	models
cvi_analyze_portfolio	Exposures, weighted scores, concentration for any holdings	portfolio
cvi_platform_status	Coverage, current research run, service health	—

Example conversation

You: Which European industrials look cheapest right now? Claude: (calls cvi_screen(sector="Industrials"), filters European indices) Fincantieri (FTSE MIB) screens cheapest — undervaluation score 0.92, Buy signal, full data confidence — per CVI research run CVIR20260715224827 (data as of 2026-07-15).

Development

pip install -e ".[dev]" && pytest -q     # offline tests, no key needed

---

Research and educational use only — not investment advice. Data licensing terms apply.