Verification as a tool call.

Agents that rebuild and migrate sites need a verification gate before shipping. MigrateKit scores the old site against the new across 7 checks and returns machine-readable JSON an agent can act on directly.

Request early access See sample JSON output

An agent that rebuilds a site cannot see what it broke.

AI tools can now rebuild a site in minutes. The problem is the gap between "the agent finished" and "the client signed off." Something dropped in that gap - a canonical tag, a piece of structured data, a form that stopped submitting - and no one knows until the rankings slip three weeks later.

A scored before/after comparison is the missing feedback loop. MigrateKit returns a structured verdict: what passed, what failed, and exactly which pages, tags, and elements failed. That is the signal an agent needs to close the loop: fix, re-run, ship when green - exactly how agents already work.

The scorecard is a gate the agent iterates against until it passes.

How agents integrate with MigrateKit

Three integration surfaces, at different stages of availability. Each card below says what exists now and what is still being built.

MCP server

EARLY ACCESS

A stdio MCP server exposing two tools: run_scorecard and run_suite. Pass two URLs, get back the scorecard JSON your agent can reason over. Works with Claude, Claude Code, and any host that speaks MCP.

What exists now: the MCP server is in the private repo and being tested with early-access partners. It is not a public package yet. Request access below and we will get you set up.

// Tool call
{
  "tool": "run_scorecard",
  "arguments": {
    "orig_url": "https://client.com",
    "dev_url":  "https://staging.client.com"
  }
}

// Response
{
  "allPass": true,
  "suites": [
    { "suite": "seo",        "score": 98, "target": 80, "pass": true },
    { "suite": "content",    "score": 100, "target": 85, "pass": true },
    { "suite": "perf",       "score": 91, "target": 80, "pass": true },
    { "suite": "a11y",       "score": 100, "target": 90, "pass": true },
    { "suite": "functional", "score": 100, "target": 90, "pass": true },
    { "suite": "responsive", "score": 100, "target": 85, "pass": true },
    { "suite": "image",      "score": 100, "target": 85, "pass": true }
  ]
}

Hosted API
IN DEVELOPMENT
A hosted REST endpoint: POST two URLs, receive the scorecard JSON. No server to run, no MCP host required. Bearer token auth, webhook support planned. Joining the waitlist now shapes the request/response spec.

What exists now: the API is in design. The JSON format is defined - you can see the exact output shape in the sample scorecard. Early-access partners who request access below will get first notice when it opens.
```
// Planned endpoint shape
POST https://api.migratekit.io/v1/scorecard
Authorization: Bearer <token>
Content-Type: application/json

{
  "orig_url": "https://client.com",
  "dev_url":  "https://staging.client.com",
  "suites":   ["seo", "content", "perf", "a11y", "functional", "responsive", "image"]
}

// Response: same JSON schema as the MCP tool output above
```
Machine-readable site
AVAILABLE NOW
Agents can read MigrateKit's pricing, capabilities, and contact point right now, without any API key. Three surfaces:
- llms.txt Plain-text summary at migratekit.io/llms.txt - tiers, capabilities, integration status in one readable file.
- JSON-LD SoftwareApplication + Offer schema on every page. Prices and descriptions match visible content exactly.
- Site check The site check is a plain POST to /api/v1/site-check with a single url field: no key, no account, no browser. It returns the detected platform, the page profile, and the risks a move would carry. Rate limited per caller.

Early access. Pricing to be decided.

Not on sale yet. Agent access is in early access while the API is validated. Partners who join early help shape the final pricing.

The Verification Report is free for a person running one rebuild, with re-runs included. That does not translate to an agent, where re-running is the entire workflow rather than an edge case. Agent access will be per run, via prepaid credit packs, so there is nothing per-seat to commit to. The packs below show the shape we are planning.

Starter pack

Early access

30 scorecard runs. Join early access and shape the pricing.
- 30 scorecard runs
- Full 7-suite JSON output per run
- All suites or individual suite calls
- No expiry on credits
Request early access
Better value

Growth pack

Early access

120 scorecard runs. Join early access and shape the pricing.
- 120 scorecard runs
- Full 7-suite JSON output per run
- All suites or individual suite calls
- No expiry on credits
- Priority queue during early access
Request early access

Early-access partners shape the final pricing. The free scan runs now - see the pricing page for details.

Request agent early access.

MCP server access goes to partners first; the API waitlist is open to anyone.