The Schema Markup Validator at validator.schema.org checks whether structured data is valid schema.org: it extracts JSON-LD, RDFa and Microdata, shows the graph it found and flags syntax mistakes, but it says nothing about Google's rich results. If Claude Code, Cursor or Codex writes your markup, the check that fits the agent loop is seogeoaeo.ai's Schema Markup Checker, a schema tool your agent can call over MCP that tests JSON-LD against Google's documented rules.
What the schema markup validator does
Schema.org's own page for the validator lists its features. It fetches a page from a URL or validates markup you paste. It reads JSON-LD 1.0, RDFa 1.1 and Microdata, combines them, and shows a summary of the extracted graph. It extracts structured data that JavaScript injects, such as markup added by widgets, and it points out syntax mistakes.
It descends from Google's old Structured Data Testing Tool and is provided by Google as a service for the schema.org community. The validator has one documented limit worth knowing: for JSON-LD it does not fetch or interpret any @context other than schema.org.
What it deliberately leaves out
When Google moved the old tool to schema.org, it said the new validator would check syntax and compliance with schema.org standards and stop checking Google Search rich result types. Google's example was a type it does not use, such as ExercisePlan: the validator is the right place for that. For anything Google turns into a rich result, Google points you to its Rich Results Test.
In practice that means markup can pass the validator and still miss what Google needs. A Product with an Offer and no price is valid vocabulary, yet it is the kind of gap a Google-focused check should flag.
Why an agent loop needs a different check
An agent that edits a layout file and adds JSON-LD needs a result it can read and act on, not a page a human scrolls. The seogeoaeo.ai checker is exposed as an MCP tool named schema-checker, with two inputs: a public URL or a raw JSON-LD string. Every finished run returns findings plus a list of tips in priority order, a status of changes, clean or incomplete, the parts it could not check, and a verify step. The agent applies the tips, deploys and runs it again until the status is clean.
Connecting it to Claude Code takes one command with a workspace API key:
claude mcp add --transport http --scope user seogeoaeo https://seogeoaeo.ai/api/mcp --header "Authorization: Bearer YOUR_API_KEY"
Each run costs 8 credits, the same as in the browser, and a call can set a credit cap so the agent never spends more than you allow.
The three checks side by side
| Schema Markup Validator | seogeoaeo.ai Schema Markup Checker | Google Rich Results Test | |
|---|---|---|---|
| Question | Is this valid schema.org | What breaks or is missing under Google's rules | Is this eligible for a Google rich result |
| Formats | JSON-LD, RDFa, Microdata | JSON-LD | JSON-LD, RDFa, Microdata |
| JavaScript markup | Extracted | Not seen, reads server HTML | Rendered |
| Agent access | Web page | MCP and REST | Web page |
| Price | Free | 8 credits per run | Free |
When to keep the validator open
The validator is still the better tool in three cases: your site uses Microdata or RDFa, you are marking up a type Google does not use, or you want to see the full graph of how your entities connect. And Google's own test remains the final word on rich result eligibility, which the seogeoaeo.ai tool page says itself.
For the routine work of writing and fixing JSON-LD inside a codebase, give the agent the checker and let it close its own loop.