Getting started

This page is the shortest path from nothing to a proposal you can approve. Six steps. Each one says what you will see on screen.

1. Create an account

Go to app.sitearc.ai and sign up. There is nothing to install first, and no API key to find.

Sitearc is hosted, so the data calls run on its infrastructure. You never see a DataForSEO key or a Google Cloud bill.

You land on the projects list. It is empty until you add a site.

2. Add a site

Click new project and enter the domain you run. A site is one registrable domain plus its subdomains. Connect a subdomain explicitly and that subdomain alone is the site.

Adding the domain starts a crawl. The crawl reads your sitemap, walks the navigation, and probes the paths a site usually has. Per page it pulls the title, the meta description, the headings and the body text. It also takes image alt text and the links in and out.

The crawl is bounded by your plan. Pages it could not reach are counted and recorded as skipped, never guessed at.

3. Connect Search Console

Open the project’s guided connect flow, or go straight to settings, then integrations. You will see two cards, one for Search Console and one for Analytics.

Click connect on Search Console and sign in with the Google account that owns the verified property. Approve read-only access. No key changes hands, and you can disconnect from the same card.

The guided flow on the integrations page handles connecting, because the production Google OAuth client is not created yet.

Do this before anything else. Search Console is a measurement source, and a change with no measurement source cannot be shipped as an experiment. The loop refuses that case by design.

Analytics connects the same way, from the card below it. Connect it too if you want session and conversion data alongside clicks.

4. Read the first report

After the crawl and your first connection, Sitearc writes the first report. It says what you have built, where the growth is hiding, and what the diagnosis is.

The report carries a confidence value and a list of missing data. Read the missing-data list first. It tells you which connection would move a conclusion from inferred to measured.

From the report you can open the research screens directly. Keywords, search performance, backlinks, competitors, rank tracking and the site audit are all live surfaces with real data.

The report page invites comments. A comment on the report is a correction, and corrections are how a wrong assumption gets fixed early.

5. Point an agent at it

Two ways, and both use the same server.

For Claude Code, install the plugin:

/plugin marketplace add ethanplusai/sitearc-app
/plugin install sitearc

The marketplace is available to account holders while the repository is private.

For any other MCP client, add the hosted server:

claude mcp add --transport http sitearc https://app.sitearc.ai/mcp

The first tool call opens an OAuth prompt. Sign in at app.sitearc.ai and approve the connection. The grant is per client, and you can revoke it from your account.

Then ask for the brief:

Read the project brief for my site and tell me what is connected.

get_project_brief is free and it is the right first call. It reports which measurement sources are connected, the active learnings, the running experiments and any recent confounders.

6. Get your first proposal

Ask the agent for a change worth making:

Using the seo-loop skill, propose one change to my site with its evidence.

The agent calls create_proposal. A proposal carries the concrete change, not an instruction to improve something. It carries at least one piece of evidence with a source and a tier. It carries a prediction: which metric, which direction, over how many days.

A proposal with no evidence is refused by the tool. A new_page proposal is refused outright.

Open the loop section of the project and go to proposals. Read the change, the evidence and the prediction, then approve or reject. Who decided is recorded, never inferred.

You can also work the queue without an agent. Each item carries the exact change and instructions to apply it yourself. Completing it that way is recorded as applied manually.

What happens after you approve

Your agent opens the pull request with its own credentials. Sitearc holds no write credential to your repository.

resolve_preview_url returns the preview deployment for that pull request. The agent audits the preview and reports what it found. Sitearc never pushes to your default branch, and you merge.

ship_experiment then records the pull request, the cohort, the holdout and the audit verdict, and collects a pre-ship baseline. Measurement runs over the predicted window. conclude_experiment computes the verdict.

What is being built

These steps are designed and underway, and none of them work yet.

• Scheduled refresh on a cadence, which will keep the report current.
• The recommendation queue, which will draft proposals after every refresh.
• The feedback tool, the content engine and the in-app chat.
• Paid checkout and credit pack purchases.
• The WordPress plugin and the GitHub App.

Everything above them works today. See each linked doc for the current state.