# Changelog

Platform and methodology releases for The MCP Census.

## July 2026 Release

### Trust scoring v3.2

- Corroboration cap — a self-declared identity earns no score without independent evidence behind it
- Brand impersonation detection across look-alike namespaces and republished catalogues
- Pillar-based scoring with every factor traceable to an evidence URL you can open
- Live endpoint verification, including OAuth-protected servers that answer a spec-conformant challenge

### Discovery beyond the registry

- Vendors that ship an MCP server without registering one are found and verified independently
- Conventional endpoint discovery across vendor documentation and API hosts
- Machine-readable manifest and `llms.txt` support for vendors publishing their own endpoints
- Every candidate clears an evidence gate before publication — nothing is listed on a claim alone

### Search precision

- Domain queries resolve on exact ownership boundaries rather than name similarity
- Match tiers separate a real name match from a passing mention
- Audited `confirmed_absent` verdicts — where no official server exists, we say so and show the review trail

### Platform

- Lookup table with frozen identity columns, universal filtering and per-column sort
- Public [Coverage](/coverage) transparency page and `/v1/coverage`
- Per-server watches with webhook and email delivery on real state transitions
- REST and MCP APIs with one-click install for major agent clients
- Multi-provider single sign-on
- [Learn](/learn) — reference material for humans and agents

Scoring rules are documented in full on the [Methodology](/methodology) page. Live totals are on [Status](/status).