API reference

Base URL

All API requests go to this root:

https://api.7275.com/v1

HTTPS is required. HTTP requests are rejected (not redirected).

Authentication

All requests need a Bearer token in the Authorization header. Generate one in dashboard → Settings → API Keys. Keys come in live and test flavors and can only access their own environment.

curl https://api.7275.com/v1/accounts \
  -H "Authorization: Bearer dxk_live_...your-key..."

• live key format: starts with dxk_live_, 64 characters
• test key format: starts with dxk_test_, scoped to the sandbox account
• Never use keys in front-end code — proxy through your own backend
• If leaked, immediately revoke and regenerate in dashboard

Use directly from Claude Desktop / Cursor (MCP server)

We ship a zero-dependency Model Context Protocol server so MCP clients like Claude Desktop and Cursor can read your AdWhiz state directly. No code to write — copy a config block, paste your API key, restart the client.

• Five read-only tools: list ad accounts, surface AI recommendations, get this-month savings, view recent activity, check API quota
• Mutation tools (apply / pause) deliberately not exposed yet — they wait until the optimizer exits shadow mode and we have calibration evidence to gate them
• Single Node 18+ script. Zero npm install. Runs on offline / airgapped machines just fine
• Your API key never leaves your machine. All requests HTTPS-direct to app.7275.com

Add this to Claude Desktop's claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%\Claude\):

{
  "mcpServers": {
    "adwhiz": {
      "command": "node",
      "args": ["/path/to/mcp-server/src/index.mjs"],
      "env": {
        "ADWHIZ_API_KEY": "dxk_live_..."
      }
    }
  }
}

Full install guide, Cursor config and troubleshooting on GitHub: github.com/dexun-inc/adwhiz/mcp-server.

Endpoints

The v1 API exposes 13 core endpoints covering identity, account reads, recommendation generation + application, audit logs, API key management, quota introspection, and the changelog feed.

Example request

Generate fresh three-platform optimization recommendations:

curl -X POST https://api.7275.com/v1/recommendations/generate \
  -H "Authorization: Bearer dxk_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "account_id": "ga-548-203-7184",
    "platforms": ["google_ads", "meta_ads", "tiktok_ads"],
    "window": "7d"
  }'

Example response:

{
  "summary": "Pausing 3 high-CPA keywords and reallocating budget...",
  "recommendations": [
    {
      "id": "rec_001",
      "type": "pause_keyword",
      "target": { "accountId": "ga-548-203-7184", "keywordText": "wool socks" },
      "title": "Pause broad-match 'wool socks'",
      "reason": "7-day CPA $45.82 vs target $8",
      "expectedImpact": "Save ~$412/week",
      "confidence": "high"
    }
  ]
}

Incrementally polling /changelog

/api/v1/changelog is a public endpoint (no API key required) with four bandwidth-saving knobs: ETag + If-None-Match, Last-Modified + If-Modified-Since (curl -z), ?since=YYYY-MM-DD for date-filtered fetches, and ?limit=N for size-capped fetches. SDKs and cron scripts can combine these to poll frequently without re-downloading the same JSON.

# 1. First call — store the ETag (or Last-Modified) you get back.
curl -i 'https://app.7275.com/api/v1/changelog?limit=10'

# 2. Subsequent calls — pass the stored ETag.
#    Server returns 304 with empty body if nothing has shipped.
curl -i 'https://app.7275.com/api/v1/changelog?limit=10' \
  -H 'If-None-Match: "abc123..."'

# 3. Or anchor on a date instead of an ETag.
#    Only entries whose `date >= since` come back; pair with ?limit
#    to cap the response size.
curl 'https://app.7275.com/api/v1/changelog?since=2026-01-01&limit=20'

# 4. curl -z for Last-Modified-style conditional GET (no ETag needed):
curl -z 'Mon, 20 May 2026 00:00:00 GMT' \
  'https://app.7275.com/api/v1/changelog'

Rate limits

Tiered by subscription plan to avoid hammering upstream Google / Meta / TikTok quotas:

• Professional: 120 requests/minute
• Business: 600 requests/minute
• Enterprise: custom (typically 3,000 requests/minute)
• Over the limit returns 429 + retry-after header — exponential backoff recommended

Error codes

All errors return JSON with error.code, error.message, and error.request_id: