# GTMRun > The week's marketing work, drafted and ready to approve. GTMRun checks a founder's website against Google's published guidance — email authentication, HTTPS, crawl files, soft 404s, conversion basics, plus a rendered-page audit, an image budget and link-graph authority from the portfolio's own tools — and turns the findings into Fix, Write and Show up cards the founder approves. This site is the preview (DRAFT, noindex, invite-only accounts): the free GTM check is live; the cards (Fix + Write) are live for invited brands; Show up cards come next. Key facts for agents: - The free check is keyless. Call the remote MCP server at /mcp (streamable HTTP, stateless, JSON responses): tools gtm_check {url, mode?}, gtm_recheck {check_id}, gtm_check_result {check_id}. GET /mcp describes it; no session, no key; about 12 check starts per hour per network. - The cards are keyed. The same /mcp server carries list_brands, list_cards {brand?, kind?, state?}, approve_card {brand, card}, skip_card {brand, card}, verify_card {brand, card} and draft_post {brand, topic?, answers?}; send the founder's sign-in token as "Authorization: Bearer " (the /app/ page holds it). Without it they answer sign_in_required. Every card is a draft until the founder approves it; a Write card's approval opens a pull request with content/blog/.md (approved: false) into the founder's repository — the merge is the publish. GTMRun never posts anywhere on the founder's behalf. - Plain HTTP for the cards: GET /api/brands, POST /api/brands {url, repo?}, POST /api/brands/{id}/run, POST /api/brands/{id}/draft {topic?, answers?}, GET /api/cards?brand=&kind=&state=, POST /api/cards/{brand}/{card}/approve|skip|verify. GET /api/config says which switches are on in this deployment. - Plain HTTP works too: POST /api/check with JSON {"url": "acme.com"} returns 202 {checkId, resultUrl}; GET /api/check/{id} returns the record; the human page is /c/{id} (real HTML, refreshes while running). - Every row is pass, fail or not measured — never an opinion score. "Not measured" is not a failure: it means the raw HTML, a catch-all host, a failed DNS lookup or an unavailable tool hid the answer; each row links to a guide that says how to verify it yourself. - Every row carries an evidence tag (google_requirement | google_recommendation | gtmrun_heuristic) and its source (own | serp | image-optimizer). Third-party measurements are labelled and are never presented as Google's view. - We follow Google's AI-search guidance: no llms.txt-as-SEO advice, no chunking, no word-count targets, no bought links. This file exists for agents reading the site, not as a ranking lever. - The GitHub Action in the repo (action/) runs gtm_check on a URL after a deploy and fails on chosen check ids. ## Docs - [Fix guides](/fix/): one guide per check — what it is, why it matters, how we check it, how to fix it, verify it yourself, sources - [How scoring works](/scoring/): the 32-point deterministic scale, informational rows, rows measured by other tools, why there is no AI score - [What we don't check](/what-we-dont-check/): llms.txt, chunking, word counts, keyword meta, bought links, third-party scores — with Google's own words - [MCP server description](/mcp): tools and connection string - [Blog](/blog/): posts written the way the Write cards write them — first-hand material, answer first, no word target - [Pricing](/pricing.html): free while in preview; prices undecided (DRAFT) - [Cards page](/app/): the founder's queue (sign-in required, invite-only) - [Changelog](/changelog.html): what changed on this site, newest first ## Optional - [Privacy](/privacy.html) · [Terms](/terms.html) · [Refund](/refund.html) · [Cancellation](/cancellation.html) · [Contact](/contact.html): DRAFT until the owner approves them - [Agentic resource manifest](/.well-known/ard.json)