Use SiteAssets from Claude Code

Model Context Protocol is a transport, not a product: it lets an agent call typed tools. SiteAssets ships one server that wraps the public API, so anything you can do in the UI you can do from a prompt.

Step 1 — Get a key

Sign in, open the dashboard and create an API key. Keys are prefixed with sk_live_ and shown exactly once — copy it straight into your secret store, never into a tracked file.

Step 2 — Register the server

Add the server to your MCP configuration. The same block works in Claude Code, Cursor, Windsurf and VS Code.

{
  "mcpServers": {
    "siteassets": {
      "command": "npx",
      "args": ["-y", "@siteassets/mcp"],
      "env": {
        "SITEASSETS_API_KEY": "sk_live_xxxxxxxx"
      }
    }
  }
}

Step 3 — Know the three tools

• check_open_graph — fetches a URL and returns the score, metadata and issues.
• suggest_open_graph_fix — returns the missing tags with recommended values and ready-to-paste HTML.
• generate_og_image — renders a 1200 × 630 image as PNG, WebP or SVG and returns its URL.

{ "url": "https://example.com" }          // check_open_graph
{ "url": "https://example.com" }          // suggest_open_graph_fix
{
  "title": "Hello world",
  "description": "Optional",
  "brandColor": "#d97706",
  "template": "minimal",
  "format": "png"
}                                          // generate_og_image

Step 4 — Ask for a fix, not a report

The useful prompt names the deliverable. “Audit the metadata on this page and open a patch” produces work; “what is my OG score” produces a number you still have to act on.

• Check https://example.com/pricing and list every failing tag.
• Rewrite the OG tags for this layout file, keeping titles under 70 characters.
• Generate a matching OG image with our brand colour and use it in the template.
• Re-check the deployed URL and tell me if the score improved.

Step 5 — Make it part of the commit

The highest-value use is a habit: after touching a layout or a page’s frontmatter, run a check. A tiny pre-push script is enough.

#!/usr/bin/env bash
set -euo pipefail
# Re-check a deployed preview before pushing metadata changes.
curl -s "https://siteassets.dev/api/v1/og/check?url=$PREVIEW_URL" \
  -H "Authorization: Bearer $SITEASSETS_KEY" \
  | node -e "let s='';process.stdin.on('data',d=>s+=d).on('end',()=>{const r=JSON.parse(s);if(r.score<100){console.error('OG score',r.score);process.exit(1)}})"

Automate the check, not the fix. The agent should explain the tag it changed.

Anonymous MCP calls are rate limited per IP; keys raise the ceiling and make runs repeatable. If you also want prompt-level recipes, the skills library documents the same workflows as reusable instructions.