# MCP Scores

> Public reliability ratings for MCP servers and x402 endpoints. Issued by MCP Authority from the MCPFax observation record. Free, anonymous, read-only.

Model version: 1.0
Public schema: 2026-08-11
Canonical site: https://mcpscores.com

## Archive coverage (measured, not advertised)

The observation archive spans 2 calendar day(s) (2026-08-11 through 2026-08-12), with observations recorded on 2 day(s).
No 30-day trend, 90-day risk ranking, survival estimate, or forward failure horizon is published, because the archive above cannot support one.

## What a score means

An ordinal 0-1000 ranking of the evidence observed for a listed endpoint. Grades: A >= 850, B >= 700, C >= 550, D >= 400, E < 400, NR = not rated.
A listing stays NR until it has at least 7 days of archive coverage and 3 successful observations. Below that floor no grade AND no numeric score is published — not in the API, not on the scorecard, not on the badge.
Every public figure comes from one code path, recomputed read-only at request time. Stored engine rating records are archive material and are never republished as a current rating.
A score is not a calibrated probability, a security audit, an endorsement, or a recommendation to transact.

## Listing identifiers

Every listing has an id: `host:<hostname>` for an x402 endpoint host, or `mcp:<publisher>/<name>` for a registry-listed MCP server.

## Free JSON API (no key)

Rate limit: per client address (cf-connecting-ip), enforced by a Durable Object counter. API routes (/api/*, POST /mcp): 60 requests per 60s. Site routes: 120 per 60s. A 429 carries Retry-After and RateLimit-* headers; if the limiter itself fails the request is served (fail open). https://mcpscores.com/health reports the enforced tiers.


- GET https://mcpscores.com/api/score?id=host:2s.io — full scorecard: rating, components, evidence, financial evidence, history, coverage
- GET https://mcpscores.com/api/top?limit=10 — current published ratings (limit 1-10). When nothing clears the publication floor this returns status "no_published_ratings" with the floor, the real archive coverage, and the per-listing reason — an accurate empty result, not an error.
- GET https://mcpscores.com/api/field — the whole register as a point field: one row per researched listing (kind, observed state, tool count, USDC treasury, observed inflow, domain year). The dataset behind the homepage network map.

## MCP server

- POST https://mcpscores.com/mcp — MCP Streamable HTTP, JSON-RPC 2.0. Free; no payment and no API key.
  Tools: score_lookup(id), top_rated(limit)
- Registry manifest: https://mcpscores.com/server.json

## Human pages

- https://mcpscores.com/ — register overview, coverage instrument, and lookup
- https://mcpscores.com/ratings — full browsable, sortable, filterable index of every researched listing
- https://mcpscores.com/insights — measured base rates for agent commerce: reachability, tenure, tool surface, money, concentration
- https://mcpscores.com/methodology — components, weights, grade bands, publication floor, hard caps, limitations
- https://mcpscores.com/badge — badge gallery in every grade, embed snippets, trust rules
- https://mcpscores.com/api — public API documentation with copy-paste examples
- https://mcpscores.com/about — MCP Authority, data sources, corrections and appeals
- https://mcpscores.com/score/<id> — full scorecard
- https://mcpscores.com/badge/<id>.svg — live 280x42 badge
- https://mcpscores.com/health — service readiness

## Corrections

Operators may dispute any observation: corrections@mcpscores.com. Include the listing id and the evidence.

## Attribution

Cite as "MCP Scores, MCP Authority, model 1.0" with the scorecard URL and the observation date shown on the card.