Language

• English

• 中文

ValueScope

A standardized DCF valuation engine your AI can call — deterministic, reproducible, and built for A-shares / HK / US / JP.

Related MCP server: ashare-mcp

What is ValueScope?

ValueScope is a standardized Damodaran FCFF DCF engine — 10-year explicit forecast, terminal value, WACC, sensitivity analysis, and reverse DCF in a fixed, reproducible framework.

Ask an LLM to "value this stock" and every conversation may use a different data source, accounting convention, and model — you can't tell whether a changed valuation means the fundamentals moved or the AI just felt different this time. ValueScope solves that with a clean division of labor:

• Your AI does the judgment — searches earnings guidance, analyst consensus, and industry benchmarks, then reasons about each assumption.

• The engine does the data and the math — A-share deducted-NI convention, non-operating-item EBIT adjustment, 10-year FCFF discounting, sensitivity, reverse DCF. Same inputs always yield the same output.

Your AI brings the intelligence; ValueScope brings the framework and the discipline. The engine itself never calls an LLM.

Supported Markets: 🇨🇳 A-shares   🇭🇰 Hong Kong   🇺🇸 US   🇯🇵 Japan

Three ways to use it: the MCP server (call it from your own AI), the web app (manual operator console), and the terminal CLI.

MCP Server

The MCP (Model Context Protocol) server lets any MCP-capable AI client — Claude, ChatGPT, Cherry Studio, Dify, and others — call the same deterministic DCF engine the web app uses. This is the recommended way to use ValueScope.

Endpoint: https://mcp.valuescope.app/mcp

Connect in two minutes

Claude Code (terminal and desktop app share one config):

claude mcp add valuescope --transport http https://mcp.valuescope.app/mcp

Claude web / mobile app: Settings → Connectors → add a custom connector, paste https://mcp.valuescope.app/mcp.

Cherry Studio and other desktop clients: add an MCP server of type HTTP with the same URL.

Then just ask in natural language: "Value Kweichow Moutai with a DCF" — no commands to learn.

How it works — one tool, two phases

get_score (four-dimension check-up) and get_relative_valuation (current multiples plus historical percentiles) answer "how is this company, and is it cheap against its own history" in a single call.

run_dcf answers "what is it worth", mirroring an equity analyst's workflow with the calling model playing the analyst:

1. Material — call run_dcf(ticker) with no assumptions. Returns the historical financials, each parameter's historical range, the engine-computed WACC and tax rate, and an analyst guide telling the model how to evaluate every assumption. No valuation is returned — a number at this point would only anchor the reasoning that follows.

2. Final — the model searches the web for guidance and consensus, reasons about each parameter, then calls run_dcf(ticker, <assumptions>) for the final valuation: intrinsic value, value bridge, forecast table, sensitivity matrix, and reverse DCF (what the market price implies).

A dcf MCP prompt is also exposed, surfacing a one-command workflow (material → search → reason → three scenarios) in clients that support MCP prompts.

Ticker format: A-shares 600519.SS / 000333.SZ, HK 0700.HK, US AAPL, JP 7203.T.

FMP key for US / JP

A-shares and HK need no key. US / JP data comes from FMP (see Data Sources). US/JP tickers get a small daily free trial served by the server; beyond that, provide your own key one of two ways:

Configure once (recommended) — pass it as a request header so every conversation uses it automatically. If you already added the server without a key, remove and re-add it:

claude mcp remove valuescope
claude mcp add valuescope --transport http https://mcp.valuescope.app/mcp --header "X-FMP-Key: YOUR_KEY"

Per-conversation — just say "my FMP key is …" in the chat; the model passes it on each call (only valid for that conversation).

Self-hosting the MCP server

The server is mounted on the FastAPI backend at /mcp (streamable HTTP). Run the backend (see Installation) and it's available at http://localhost:8000/mcp. Set FMP_API_KEY in the environment to enable the US/JP trial; tune MCP_DAILY_LIMIT and MCP_US_TRIAL_DAILY_LIMIT for rate limits.

Web App

Try it at valuescope.app — no installation required.

The web app is the manual operator console: dial in DCF parameters by hand, watch the valuation update live, read sensitivity tables, and eyeball relative-valuation percentiles and multi-factor scores. If you like driving the assumptions yourself, it's a solid DCF calculator.

Features

• DCF Valuation — Damodaran FCFF framework with interactive parameter controls, 10-year forecast table, dual sensitivity analysis (Growth×Margin, WACC), and bridge-to-value breakdown.

• Relative Valuation — Current multiples (PE, PB, PS, EV/EBITDA) vs historical percentiles across 3/5/10-year windows.

• 4-Dimension Scoring — Valuation, Quality, Growth, and Momentum in a radar chart with transparent sub-factor breakdown.

• Financial Overview — Key drivers (revenue growth, EBIT margin, ROIC, FCF), balance sheet highlights, and historical financial table.

• Bilingual UI — English and Chinese with one-click toggle.

Terminal CLI

For local use with your own AI CLI subscription. Requires Python 3.8+.

• AI Copilot — local AI engine (Claude / Gemini / Qwen) suggests parameters; you review and adjust interactively.

• Custom Valuation — full manual control with --manual. No AI or API key required.

• Auto Mode — fully automated with --auto: AI → accept → export Excel.

• Excel Export — formatted .xlsx with valuation results, historical data, and AI reasoning.

Auto-detects installed engines (priority: Claude > Gemini > Qwen), or force one with --engine. If none is found, falls back to manual mode.

Data Sources & FMP API Key

💡 Get an FMP API Key →

FMP (Financial Modeling Prep) provides high-quality financial data for US, HK, and JP markets. Subscribing through this link (coupon valuescope included) is discounted — and supports ValueScope's ongoing development.

Installation & Usage

Option 1: MCP Server (Recommended)

No installation — connect your AI to https://mcp.valuescope.app/mcp (see MCP Server above).

Option 2: Web App

Visit valuescope.app — no installation needed.

Option 3: Self-host (CLI + backend + MCP)

Requires Python 3.8+.

git clone https://github.com/alanhewenyu/ValueScope.git
cd ValueScope
pip install -r requirements.txt          # CLI
pip install -r requirements-api.txt      # backend + MCP server

Set your FMP API key (required for US/Japan):

export FMP_API_KEY='your_api_key_here'

Run the CLI:

python main.py                    # AI copilot (default)
python main.py --manual           # Manual input
python main.py --auto             # Fully automated

Or run the backend (serves the REST API and the MCP server at /mcp):

uvicorn backend.main:app --host 0.0.0.0 --port 8000

Architecture

valuescope/
├── frontend/              # Next.js (React) — web UI
├── backend/              # FastAPI — REST API + MCP server
│   └── mcp_server.py     # MCP tool (run_dcf) + dcf prompt
├── modeling/             # Core valuation engine (shared by CLI, backend, MCP)
├── main.py               # Terminal CLI entry point
└── Dockerfile            # Backend container

The modeling/ engine is the single source of truth — the CLI, the web backend, and the MCP server all call it, so a valuation is identical no matter how you reach it.

Key Valuation Parameters

Contributing

Issues and pull requests are welcome. Contact: alanhe@icloud.com

For more on company valuation, visit jianshan.co or scan to follow on WeChat:

License

This project is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0).

You are free to use, modify, and distribute this software, but any modified version — including use as a network service (SaaS) or a hosted MCP server — must also be open-sourced under AGPL-3.0.

© 2025-2026 Alan He