# CONVERG3NCE

CONVERG3NCE audits a brand's web presence the way an AI shopping agent actually reads it,
scores how agent-ready the brand is, and says what to fix.

As LLMs answer "what should I buy" directly, brands become invisible to a channel they
cannot see or control. These docs describe the machinery that measures that channel:
the audit pipeline, the Agent-Readiness score, and the public MCP server an agent can
query directly.

## The loop

```
AUDIT ──▶ DIAGNOSE ──▶ SCORE ──▶ FIX ──▶ DEFEND
  │           │          │        │        │
  │           │          │        │        └─ rerun monthly, store everything
  │           │          │        └────────── prioritized fixes, ranked by impact
  │           │          └─────────────────── deterministic 0–100, no LLM
  │           └────────────────────────────── LLM judge reads each response
  └────────────────────────────────────────── query 4 model families with tailored questions
```

## Start here

[!ref Target=_self text="How the audit works"](audit/)
[!ref Target=_self text="The Agent-Readiness score"](audit/scoring.md)
[!ref Target=_self text="Connect to the MCP server"](mcp/)

## What is documented, and what is not

Everything on this site is derived from the running system — the audit engine's own
source, its scoring arithmetic, and measurements taken against real brands. Where a
number is uncertain, the uncertainty is stated rather than smoothed over. See
[Measurement reliability](audit/reliability.md) for the honest limits of the score.

This site does not document internal operations, client engagements, or the operator
dashboard. Those are not public surfaces.

!!!warning Access
The MCP server and the `/v1/verify` endpoint are API-key gated. There is no self-serve
key issuance — contact CONVERG3NCE for a key. See [Authentication](mcp/index.md#authentication).
!!!