You don't always want to write a script. Sometimes you just want to ask: "What's Keyence's revenue CAGR over 5 years?" or "Compare Toyota, Honda, and Nissan side by side" and get an answer.
That's what MCP does. It gives Claude direct access to the Axiora API — 19 tools covering financials, screening, rankings, growth, health scores, and filing translations. You ask in English (or Japanese). Claude calls the right tool, gets the data, and answers.
Setup takes 3 minutes. No SDK, no Python, no code at all.
What Is MCP?
Model Context Protocol is an open standard that lets AI assistants call external tools. Instead of pasting data into a chat window, Claude connects to the data source directly and pulls what it needs.
Axiora's MCP server exposes 19 tools:
| Tool | What it does |
|---|---|
search_companies | Find companies by name (JP/EN), ticker, or EDINET code |
search_companies_batch | Look up multiple companies at once (mix of codes and names) |
get_company | Company details — sector, listing, fiscal year end |
get_financials | Revenue, net income, assets, equity, ratios — up to 20 years |
get_growth | YoY growth rates and 3/5-year CAGRs |
get_ranking | Rank by any metric (ROE, revenue, margin, etc.) |
screen_companies | Filter by sector, min revenue, min ROE, max P/E |
compare_companies | Side-by-side financials for 2–5 companies |
get_peers | Find similar companies by sector and revenue proximity |
get_health_score | Financial health (0–100) with component breakdown |
get_health_ranking | Rank companies by health score |
get_timeseries | Chart-ready time series for any metric across companies |
get_sector_overview | Sector stats and company counts |
search_translations | Full-text search across English translations of filings |
get_translations | Read translated sections of a specific filing |
list_filings | Browse filings by company, type, or date |
get_filing_calendar | Filing volume by day for a given month |
get_coverage | Data coverage stats — totals, metric availability, freshness |
get_documentation | API docs, metric definitions, and usage guide |
get_documentation is worth noting — Claude calls it on first connect to learn what's available, what each metric means, and how the tools fit together. The server is self-documenting. No prompt engineering required on your end.
All data comes from audited EDINET filings (有価証券報告書), XBRL-parsed.
Setup: 3 Minutes
You need two things: an Axiora API key (free, no credit card) and Claude Desktop.
Add this to your Claude Desktop config:
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"axiora": {
"url": "https://api.axiora.dev/mcp",
"auth": {
"type": "bearer",
"token": "YOUR_API_KEY"
}
}
}
}Restart Claude Desktop. You'll see the Axiora tools listed in the tool picker. That's it — no Docker, no local server, no dependencies.
Conversation 1: Company Deep-Dive
You: "What are Keyence's financials for the last 5 years?"
Claude calls get_financials with code 6861 and years=5, then formats the response:
Keyence (6861) — 5 Year Financials
Year Revenue Net Income ROE Op. Margin 2024 ¥2.4T ¥1.0T 19.5% 54.2% 2023 ¥2.2T ¥938B 18.3% 52.8% ... Operating margins consistently above 50%. ROE stable at ~19%.
You: "What's their revenue growth rate?"
Claude calls get_growth — gets YoY rates and CAGRs:
5-year revenue CAGR: 15.2%. Revenue has grown every year except 2020.
No context switching. No opening a new tab. The conversation builds on itself.
Conversation 2: Peer Comparison
You: "Compare Toyota, Honda, and Nissan's financials for the last 3 years."
Claude calls compare_companies with codes ["7203", "7267", "7201"]:
Toyota Honda Nissan 2024 Revenue ¥29.9T ¥16.9T ¥12.7T 2024 Net Income ¥4.9T ¥1.1T ¥426B ROE 15.2% 10.8% 5.1% Toyota's net income is larger than Honda and Nissan's combined. Nissan's ROE at 5.1% signals profitability challenges.
You: "Who are Toyota's closest peers by revenue?"
Claude calls get_peers with code 7203:
Top peers in 輸送用機器 (Transport Equipment) by revenue proximity: Honda, Denso, Suzuki, Subaru...
Conversation 3: Screening
You: "Find Japanese companies with ROE above 20% and revenue over ¥500 billion."
Claude calls screen_companies with min_roe=20.0 and min_revenue=500000000000:
Found 18 companies matching your criteria:
- Keyence (6861) — ROE: 19.5%, Revenue: ¥2.4T
- Tokyo Electron (8035) — ROE: 25.3%, Revenue: ¥2.2T
- ...
You: "Now show me the top 10 electronics companies by operating margin."
Claude calls get_ranking with metric operating_margin and sector 電気機器:
- Keyence — 54.2%
- Lasertec — 44.8%
- ...
You're doing multi-step financial analysis in a chat window. Each answer takes seconds.
Conversation 4: Reading Filings in English
You: "Search filings that mention semiconductor risks."
Claude calls search_translations with query semiconductor:
Found mentions in filings from Tokyo Electron, Renesas Electronics, Advantest... Snippet from Tokyo Electron's MDA: "...semiconductor equipment demand is cyclical and subject to..."
You: "Show me Tokyo Electron's full risk factors section."
Claude calls list_filings to find the doc ID, then get_translations with section risk_factors:
The filing's risk factors section, translated to English, covering supply chain dependencies, currency exposure, and geopolitical risks.
This is genuinely useful. Japanese filings are in Japanese. The translations make them accessible to anyone.
What Works Well
Multi-step analysis. Ask a broad question, then drill down. Claude remembers context — "now show me their growth" refers to the company you were just discussing.
Natural language screening. "Companies in the IT sector with ROE above 15%" becomes a screen_companies call automatically. No syntax to learn.
Cross-tool queries. "Compare Toyota and Honda, then show me Toyota's peers" chains multiple tools in one response.
What Doesn't (Yet)
No real-time market data. This is filing data — annual reports, audited financials. No stock prices, no market cap, no intraday anything.
No charts in Claude Desktop. Claude can describe trends, but you can't render a line chart in the chat window. For visualization, use the Streamlit screener or the SDK with matplotlib.
Sector names are in Japanese. EDINET classifies sectors in Japanese (電気機器, 情報・通信業). Claude handles this fine — you can say "electronics sector" and it maps to 電気機器 — but it's worth knowing.
Keep Going
- Japanese Financial Data with Python — if you want to write code instead
- Build a Japan Stock Screener with Streamlit — a full UI for screening
Free tier. No credit card. Get your API key →