Delivery model

The report is a server,
not a PDF.

In plain terms: instead of emailing you a PDF nobody opens twice, we give you a connection you add to Claude or ChatGPT. Then anyone on your team can ask “what changed on our site this month?” and get an answer, without contacting us and without logging into another dashboard.

A read-only version of this would be a report viewer with extra steps. So the MCP server is the primary delivery mechanism: your CEO asks it questions in Claude, your developers point their coding agent at it and let it apply the fixes.

Scan first

Connect in about thirty seconds

Try it against demo data now.

The public sandbox is read-only and serves fixtures for three fictional sites. No account, no key. It exposes the same tool surface your production connection will.

1

Copy the endpoint

https://mcp.code4x.com/sse

2

Add it as a connector

In Claude: Settings → Connectors → Add custom connector. In Cursor: add it to mcp.json in your project.

3

Ask it something

“Show me the readiness score for northgate.io and what changed since the first run.”

mcp.json

{
  "mcpServers": {
    "code4x": {
      "url": "https://mcp.code4x.com/sse"
    }
  }
}

cursoragent session

> Audit staging and fix what you can

● run_audit staging.northgate.io
  → 11 failures, 4 auto-fixable
● get_remediation L1-01, L1-04, L1-13, L2-01
  wrote public/llms.txt
  wrote public/agents.json
  wrote middleware.ts
  patched public/robots.txt
  4 files changed · review the diff

The audit stops being a report someone reads and becomes a task an agent resolves.

Tool surface

Seven tools.

run_audit(domain, max_pages, rubric_version) Triggers a run. Returns a run_id immediately; completion arrives by webhook.

get_score(run_id) Headline score, per-layer breakdown, per-dimension scores, and the evidence for every lost point.

compare_runs(domain, run_a, run_b) Per-dimension deltas. Highlights regressions as prominently as improvements.

list_flags(run_id, severity, dimension) Tier 2 advisory findings with evidence snippets, URLs, and the pinned model version that produced them.

get_remediation(run_id, dimension_id) Generated fix files for a failed check — populated, not templated.

get_representation(domain, period) Citation sources, accuracy verdicts, sentiment and engine breakdown for the period.

generate_report(run_id, format) Client-facing HTML or PDF. Returns a download URL.

Sandbox restrictions: read-only fixtures, no run_audit, no get_remediation writes, rate-limited per connection.

Who asks what

One connection, four very different conversations.

Persona	What they ask	What comes back
CEO	What's our score vs. last month?	get_score + compare_runs
CTO	What's critical on checkout?	list_flags(critical, /checkout)
Developer	Audit staging and fix what you can	run_audit + get_remediation
Marketing lead	How are AI engines describing us?	get_representation