SeoGrep

We ran SeoGrep on seogrep.com. Here is what it found.

An SEO tool audited by itself: what our crawl, on-page, and schema checks found on seogrep.com, and the fixes that shipped.

SeoGrep exists to answer one question about a website: what's broken, and what's worth fixing first. The most direct way to check whether that answer is any good is to ask it about a site we control completely — our own. On 2026-07-20 we ran SeoGrep against seogrep.com, start to finish, the same way any user would: from a chat window, in plain language.

Setting up the run

We didn't reach for a database console or a one-off script. We opened a chat with SeoGrep connected over MCP and asked for a project and a crawl — the same two tools, setup_project and crawl_site, that any new user's first session runs. crawl_site is asynchronous: it hands back a job_id and a background worker does the fetching, following the sitemap and same-origin links and respecting robots.txt, so the next step was just polling get_job_status until it reported done.

The crawl covered 42 pages, comfortably inside the current 100-page-per-run cap.

Running the audits

With a crawl on record, two synchronous tools did the actual checking: audit_onpage for titles, meta descriptions, headings, and canonicals, and audit_schema for structured data. Both return their findings directly in the chat, no dashboard or export required.

Here's roughly what that looks like in a client. This transcript is illustrative — paraphrased to show the shape of the exchange, not a verbatim log:

You: Run an on-page audit for my seogrep.com project.

SeoGrep: 42 pages checked.
  - Canonicals: 42 missing, 0 present
  - Meta descriptions: 5 pages share one duplicate description,
    several others run past the recommended length

You: Now check structured data.

SeoGrep: 0 of 42 pages carry any JSON-LD. No Organization,
no WebSite, nothing for a crawler to key off.

The real findings were just as blunt as that reads:

  • Canonicals — 42 of 42 pages had no canonical tag at all. Zero.
  • Structured data — audit_schema found 0 of 42 pages with any JSON-LD structured data.
  • Meta descriptions — five marketing pages shared the exact same 178-character description: duplicate content, and past the usual display length in one move. Several docs pages ran well over 155 characters too.
  • Technical health — the one clean result. audit_tech showed all 42 pages returning a 2xx status, no redirect chains, and no robots conflicts. The plumbing was fine; the metadata layer was not.

That's a rough scorecard for a tool that sells SEO audits. Which was exactly the point of running it.

What shipped afterward

Findings that stay findings aren't worth much, so here's what changed on the back of this run:

  • Every route now carries a self-referencing canonical tag.
  • Meta descriptions are unique per page — no more five pages sharing one string.
  • The marketing pages carry Organization, WebSite, and SoftwareApplication JSON-LD.
  • Docs meta descriptions are truncated to 155 characters or fewer, cut at a word boundary instead of mid-word.

None of that required touching the audit tools themselves — audit_onpage and audit_schema didn't change. The site had to catch up to what they were already checking for. For the tools-reference pages specifically, the description-length fix isn't a one-off hand-edit either: the docs generator enforces the 155-character cap and truncates at a word boundary itself, so a future tool description can't quietly push a page over the limit again.

What we are not claiming

We are not going to tell you canonical tags and JSON-LD moved seogrep.com's rankings, because we haven't measured that — no before/after traffic number, no ranking chart, nothing to point to. If we had one, we'd publish it here with the same directness as the 42 out of 42.

What we can say is narrower, and we think more useful: an SEO tool ought to survive being pointed at itself. Ours found real, unglamorous, fixable problems in its own marketing pages and docs, and fixing them was mechanical work — canonical tags, unique descriptions, structured data, description length — exactly the kind of thing the tool exists to surface. If you want to see what it finds on your own site, setup is a personal MCP URL pasted into your client.

← All posts