DT 1.5 runs two ways: the dashboard in your browser, and mgo, the terminal CLI. Same agent, same approval gate, your call. Both tracks are below with the exact steps and copy-paste commands.
They run the same audit engine and the same approval loop. The dashboard explains everything on screen; the CLI is for people who live in a terminal, the way Claude Code and Codex do it.
Open it, type your URL, and read the findings in plain language. Drafted fixes wait in an approval queue. Nothing to install.
DT 1.5 in your terminal: an interactive agent, audits, a deterministic plan, copy-paste JSON-LD, a real approval loop, and AI scans on your own keys.
Four steps, about two minutes. The audit is a live crawl of your real pages, and it is labeled on screen if the live service is unreachable and a recorded sample is shown instead.
Go to the free DT 1.5 demo. It opens on the overview and a short guided tour walks you through every part of it, step by step, so you can skip it or follow along. Everything is in the left navigation: overview, task queue, approvals, the audit runner, and the ad manager. That page is the free demo; the full dashboard, where your account, queue, and connections are saved, is at mgodata.com/dashboard. You can reopen the tour any time from the Help button at the bottom of the left navigation.
Accounts and sign-in are live. In the full dashboard, create an account; no payment is required during the current open beta. Your queue persists between visits and Connections lets you connect Shopify, Search Console, an AI provider, and supported Google Ads, Meta, or TikTok ad accounts. Ad Manager verifies the exact account assets before approval; activation remains a separate decision. Google Business Profile remains unavailable while MGO awaits Google's separate API approval.
Click Run Audit in the left navigation, type your site's address, and press Run audit. Findings come back ranked by severity, written in plain language. To see a real recorded audit of our own store instead, type:
purivial.comWhere the audit finds a supported issue, DT 1.5 drafts the patch. In your dashboard, queue a suggestion, open Approvals, and read the exact change before it goes anywhere. Approve it or reject it; nothing ships without your yes. In the demo the queue is a read-only preview. The Shopify SEO automation page explains which approved changes can be verified on a connected store and which remain paste-ready patches.
The package is @mgodata/mgo and the command is mgo. Deterministic audits and fixes need no AI keys; they unlock with your signed-in MGO Data account, including during the current open beta. CLI AI lanes use the provider keys you bring, sent provider-direct.
You need a terminal, Node.js 22 or newer, your website's address, and an MGO Data account. The audit and fix commands unlock after mgo login; open-beta access requires no payment.
One global npm install.
npm install -g @mgodata/mgomgo login is a browser device flow, like the gh or claude CLIs: no password is typed into the terminal, and the real commands unlock after sign-in. mgo doctor checks your Node version, config, local state, and whether the audit service is reachable. Provider keys are reported present or absent only; their values are never printed or stored.
mgo --version
mgo login
mgo doctorA live crawl of your homepage, robots.txt AI-crawler rules (GPTBot, ClaudeBot, PerplexityBot, and more), sitemap.xml, llms.txt, Open Graph, and JSON-LD. Findings print ranked by severity. Every crawl is cached locally so later commands work offline.
mgo audit yourstore.commgo plan ranks the real findings into a prioritized worklist with a fixed, documented scoring table. mgo schema prints the copy-paste-ready JSON-LD blocks. Both are deterministic: same crawl in, same output out, no language model involved.
mgo plan yourstore.com
mgo schema yourstore.comQueue each drafted fix as an action request, approve or reject it, then apply the approved patch to a local HTML file. The file is re-read from disk and the action is only marked verified when the approved block parses back out. Every state change lands in an append-only audit log.
mgo schema yourstore.com --queue # file each suggestion for approval
mgo approvals # list what is waiting on you
mgo approve <id> # or: mgo reject <id>
mgo apply <id> --file index.html # write the approved patch, verify itAction types that need platform approval or keys (ads, gbp) refuse to apply, print the honest reason, and exit with code 2. They are never faked.
mgo ai-scan asks AI answer engines about your brand and prints their real responses. It runs on your own API keys, read from your environment and sent straight to the provider you pay. They never touch an MGO server. One env var per lane:
| Env var | Powers | Notes |
|---|---|---|
| ANTHROPIC_API_KEY | Anthropic Claude lane | model knowledge, no live web in this build |
| OPENAI_API_KEY | OpenAI lane | model knowledge, no live web in this build |
| PERPLEXITY_API_KEY | Perplexity Sonar lane | web-grounded, cites sources |
export ANTHROPIC_API_KEY="sk-ant-..." # set whichever lane's key you have
mgo ai-scan "Your Brand"Set several keys and pick which engine answers first with mgo config set default_provider anthropic (or openai, or perplexity). Honest caveat: API answers approximate the consumer apps, they do not exactly equal them. Treat scans as directional.
The same keys also power the LLM drafting lanes and mgo director. Where a URL is given, each draft is grounded in the same live crawl as the audit, filed into the approval loop as pending, and never auto-published. After approval, the meta and llms.txt drafts apply to a local file with mgo apply; the Director writes an advisory weekly plan built from your real findings and queue.
mgo draft meta yourstore.com # better title + meta description
mgo draft llmstxt yourstore.com # curated llms.txt (optional; not a citation lever per Google)
mgo draft adcopy yourstore.com # ad headlines + primary texts
mgo draft reply "<review text>" # review reply, in your brand voice
mgo draft voice yourstore.com # short brand-voice guide
mgo director yourstore.com # advisory weekly plan, never publishesHonest caveats, straight from the CLI's own help: drafts are never auto-published, and publishing for ad copy, review replies, and brand voice stays manual because the ad-platform and Google Business Profile write APIs are review-gated. Without a key these commands say so and exit with code 2.
The CLI's whole surface: run mgo --help in your terminal for the same list. Anything not built or gated by a platform review says so when you run it and exits with code 2. Nothing fakes success.
| Command | What it does | Needs |
|---|---|---|
| mgo login / whoami / logout | Browser device-flow sign-in to your MGO Data account, like the gh CLI; real commands unlock after sign-in | account · open beta |
| mgo audit <url> | Full AI-visibility audit from a live crawl | account · open beta |
| mgo plan <url> | Deterministic prioritized worklist from the same crawl | account · open beta |
| mgo schema <url> | Copy-paste-ready JSON-LD blocks; --queue files them for approval | account · open beta |
| mgo approvals / approve / reject | The approval queue and its state machine | account · open beta |
| mgo apply <id> --file f | Write an approved patch into a local HTML file, then verify it | account · open beta |
| mgo log / status / doctor / config | Audit trail, local snapshot, health checks, settings | no account needed |
| mgo connect <provider> | Store a platform secret (Shopify token, AI key) encrypted server-side, never plaintext | account · open beta |
| mgo execute | Drain approved fixes through the executor: Shopify write or local file, then re-audit to verify | service-role setup |
| mgo (bare) / mgo agent | The conversational AI agent: plain language over every lane, every tool call shown | your AI key |
| mgo ai-scan <brand> | What AI answer engines say about you, real responses | your AI key |
| mgo draft <kind> | LLM drafting lanes (meta, llms.txt, ad copy, review reply, brand voice), filed for approval, never auto-published | your AI key |
| mgo director <url> | LLM weekly plan from real findings, advisory only | your AI key |
| mgo login --supabase | Self-host the control plane: the approval loop on your own Supabase project | your Supabase project |
| mgo seo report / connect | Designed, not built yet; running them says so | rolling out |
| mgo gbp / ads | The standalone CLI shortcuts remain gated. Connected ad setup and approval-gated execution live in the dashboard Assistant and Ad Manager; Google publishing still depends on Google's approval. | approval gated |
Start in the browser now, or take the CLI track now. And if you would rather hand it all to people, that door is open too.