stox-mcp
Give your AI the numbers that matter. An MCP server over the
stox.market metric layer — context, not data: every
finance feed hands a model raw bars and statements, and the model still can't
answer "is this number good?". This server ships the judgment layer instead.
Live endpoint: https://mcp.stox.market/mcp (streamable HTTP) ·
Demo & docs: stox.market/mcp
Tools
Design
- Token economy — compact JSON, score + band + a one-line read; a
fingerprint is ~550 tokens where a statements dump is 20,000.
- Percentiles, not raw numbers — every score is a cross-sectional rank
against the real cohort (100 = best), so "is it good?" is answered by
construction.
- Provenance — snapshot dates on scores, links back to the full picture.
- Informational only — descriptive statistics; never investment advice,
no buy/sell recommendations. Derived metrics only; this is not a
market-data feed.
The human-readable twin
The same metric layer renders as editorial leaderboards at
stox.market/lists — 31 boards plus sector
slices and frozen earnings-season editions, refreshed on automated weekly
lanes. A board is where a name catches your eye; fingerprint is how your
agent vets it in one call.
Universe: S&P 500 + NASDAQ-100 + curated ETFs. Coverage and metric
definitions: stox.market.