Give your AI the numbers that matter.
Every finance data feed hands your model raw material — bars, quotes, statements — and the model still can't answer "is this number good?". The Stox MCP server ships context, not data: cross-sectional percentile fingerprints, curated competitive peers, co-movement themes, and market-stress state. Compressed for context windows; provenance on every answer.
Connect it
Works with any MCP client over streamable HTTP. Informational only — it describes, ranks, and cites; it never gives investment advice.
Then ask things like:
- What's the character of MU right now?
- Compare Costco to its actual competitors.
- Who does NVDA trade with, versus who it competes with?
- Any energy names strong on both price and fundamentals?
- How stressed is the market today?
Try it right here
This calls the live endpoint from your browser and shows the exact JSON a model receives — with the latency and the token count, because the size of an answer is part of its quality.
the character of one stock. 16 percentile scores vs the S&P 500 cohort, the P3 composite (the validated 3-factor form), curated competitive peers, and a one-line read — ~550 tokens.
What's distinctive
- Percentiles, not raw numbers. Every score is a rank against the real cohort — the "is 23% gross margin good for a semi?" question answered by construction.
- True peers. Costco's competitors are Walmart, Target and Kroger — not "Discount Stores". Curated from segment overlap, per company.
- Token economy. A fingerprint is ~550 tokens where a statements dump is 20,000. Small answers are a feature, not a limitation.
- Provenance. Snapshot dates on scores, sources and dates on verdicts, and a link back to the full picture on stox.market.
Derived metrics only — this is a judgment layer, not a market-data feed. Descriptive statistics and research citations, never investment advice.