Yes, through MCP. Point any MCP client at:
https://app.ranksmile.pl/mcpThe server speaks Streamable HTTP and authorizes with OAuth 2.1. Your client registers itself, you approve it in the browser once, and the token it receives is scoped to your account. No API key to paste anywhere, and nothing to rotate by hand.
What the assistant can do
Twelve tools; eleven only read:
| Tool | What it returns |
|---|---|
workspace__list | Your workspaces and their primary domains |
workspace__get_context | Brand name and notes, voices, competitors found in search, and the research log |
article__list | Articles with stored scores and word counts |
article__get | One article's metadata, scores and optionally its body |
article__score | The per-slot breakdown behind a content score |
article__optimize_log | Auto-Optimize history: score and length before and after each run |
article__jobs | Recent generation jobs, with what the generator was told and what it returned |
visibility__overview | How often AI answers name your brand, and which brands they name instead |
rank__keywords | Google positions of your tracked keywords |
gsc__performance | Last 30 days of Search Console clicks, impressions, CTR and position |
audit__overview | The last site crawl: health scores and each issue with the pages it affects |
workspace__update_context | Adds to or rewrites your brand notes and adds to the research log |
What it can change
Only your brand notes and research log, and only if you leave Read and write selected when you approve the client. Choose Read only and it can't change anything.
An assistant reading your scores is useful. An assistant rewriting your published articles because it misread an instruction is a bad afternoon. No tool edits, publishes or deletes articles: that stays in the app, where a person approves it.
What it's actually good for
The honest answer is debugging. article__score is the tool that says why a piece scores badly: which terms are under-used, how the length compares to the competitor-derived target, which structural slots are empty. Ask an assistant to explain a bad score and it can read the breakdown instead of guessing from the text.
article__optimize_log covers the other common frustration. When Auto-Optimize appears to have changed nothing, the log holds the before and after scores and the rejection reason, which is usually the whole explanation.
Setting it up
The MCP page has per-client configuration, including Claude Desktop, Claude Code and Cursor. Most clients need one entry in a config file and a browser approval.
One connection, one account
The token is tied to the account that approved it, and the assistant sees exactly the workspaces that account can open. Connecting a client does not widen anyone's access.
Need help?
Write to us and a person answers. See the frequently asked questions first if your question is a common one, or get in touch for anything else.