# Claude Code: your AI visibility in the terminal you work in > For the complete documentation index, see [llms.txt](https://attensira.com/llms.txt). > Every page on this site is also served as Markdown: append `.md` to any URL, or send `Accept: text/markdown`. > Add Attensira's MCP server to Claude Code with one command, sign in from /mcp, and read, approve and hand off SEO work without leaving the terminal. - Canonical page: https://attensira.com/integrations/claude-code - Family: integrations - Last updated: 2026-10-04 - Author: Karl-Gustav Kallasmaa (https://x.com/kallasmaa) - Claims as data: https://attensira.com/claims.json?slug=claude-code **In short:** Add the Attensira server to Claude Code with one command, then sign in from /mcp. Claude Code can then read which prompts you lose and who wins them, approve what the crew has waiting, and hand it new work, in the terminal where you already are. No key touches a config file, and reading is free. ## What it showed on attensira.com (read 2026-10-04) Asked in Claude Code: "Which tracked prompts are we losing worst this week, and who wins them?" Call: `get_analytics area=prompt days=7 order=worst_first` We are named in none of the answers to these four, and a different rival leads each one. | Row | Value | Note | | --- | --- | --- | | "AI crawler logs analysis tool" | 0 of 42 | Ahrefs named in 14 of 42 | | "AI prompt volume tool" | 0 of 36 | Profound named in 16 of 36 | | "AI search visibility API" | 0 of 24 | SE Ranking named in 17 of 24 | | "AI search visibility consultant" | 0 of 42 | Semrush named in 3 of 42 | _Four of the six worst over seven days, ChatGPT and Perplexity._ ## What can Claude Code do between commits? - Read which tracked prompts you lose, who is named instead, and the verbatim answers behind each rate, in the terminal where you are already working. - Check whether a page you are about to change is cited at all, and on which engine, before you touch it. - Approve or decline the crew's waiting work, or hand it new work as plain lines with ask, without opening the dashboard. - Script the same reads in a headless job with a workspace API key, when nobody is there to sign in. ## How do I connect Claude Code? 1. **Add the server with no credential** Run claude mcp add with the http transport and the server URL. Nothing secret is supplied here, which is the reason to prefer OAuth in Claude Code. Add --scope user to make it available in every project. 2. **Sign in from /mcp** Run /mcp, select attensira and choose to authenticate. A browser opens, you sign in and approve, and the callback returns to a local port Claude Code listens on. Two instances can sign in at once, each on its own port. 3. **Check the connection** Run claude mcp list, then ask Claude to call get_account. It lists the workspaces the connection reaches, the credit balance, the scopes and the platforms really measured. 4. **Ask in your own words** You do not need tool names. Ask which prompts you lose, which pages get cited or what is waiting for approval, and Claude Code picks the call. Reads are free. Only tools that start agent work, such as ask, run_automation and add_prompts, spend credits; the MCP tool reference lists every one. ## What won't Claude Code do with it? - A signed-in connection always carries read and write. There is no read-only consent, so Claude Code can add prompts, run automations and spend credits if you ask it to. For reads only, use a read-only API key. - It cannot tell you why a model changed its answer. A rate tells you what changed; the answer text from answer_history is where any reason lives, and a provider-side change looks the same as a content problem. - A week of readings is thin. On Starter each prompt is read at most once per engine per day, so a seven-day rate rests on single-digit n per engine. ### What did Claude Code show on our own workspace? Four losing prompts, four different winners. That is four separate briefs, not one "visibility problem", and it took one call to see it, mid-task, without opening a dashboard. The next sentence in the same terminal is which of the four to hand the crew first. ### Why is a terminal the right place to ask? The question "are AI answers naming us?" usually comes up while the site is changing, and the cost of answering it elsewhere is a context switch nobody makes. In Claude Code the useful loop is four sentences: which prompts are we losing, read me the worst answer, what decides it (`win_plan`, free), hand that to the crew (`ask`, which spends credits). Two Claude Code instances can sign in at the same time, each on its own local callback port, so a second terminal does not log the first one out. ### Where does work handed off from the terminal land? Not as a diff in your terminal. It arrives where your site lives: a pull request on your repository if GitHub is connected, a draft in Webflow, WordPress, Shopify or Notion, or text to paste if nothing is connected. You review it there, or approve it from Claude Code with `approve_work_item`, under the workspace's approval mode. ## FAQ ### Do I need an API key for Claude Code? No. Add the server with no credential and sign in from /mcp. A key is for machines without a browser, such as a CI job, and then goes in an Authorization header read from an environment variable. ### How do I make Attensira available in every project? Add --scope user to the claude mcp add command. Without it, the server is added to the current project only. ### Can Claude Code reach more than one workspace? Yes. The connection reaches every workspace your organisation owns, and every workspace tool takes an optional project. get_account lists the ids; name the workspace in your request when there are several. ### Can I run two Claude Code sessions at once? Yes. Each instance listens on its own loopback port during sign-in, and both are accepted. ### What will Claude Code do without asking me? Only what you ask it to, with the tools you allow. The crew's own work still follows your workspace's approval mode, Draft by default, and anything stating a price, a number or a competitor's name waits for a person. ## Sources Each entry below is one claim on this page. `retrieved` is the date the exporter or a human last saw the source say this, and nothing more: it is not a publication date and not a promise of re-verification. `cite` is the anchor on the canonical page where this claim is rendered. ### claude-code-two-instances Attensira's connect documentation states that two Claude Code instances can authenticate at the same time because each listens on its own ephemeral loopback port. - source: https://docs.attensira.com/mcp/connect (docs.attensira.com) - retrieved: 2026-10-04 - cite: https://attensira.com/integrations/claude-code#claim-claude-code-two-instances > Two Claude Code instances can authenticate at the same time. Each listens on its own ephemeral loopback port, and both are accepted. ### oauth-read-and-write Attensira's tool reference states that an OAuth connection always carries read and write and that there is no read-only consent. - source: https://docs.attensira.com/mcp/tools (docs.attensira.com) - retrieved: 2026-10-04 - cite: https://attensira.com/integrations/claude-code#claim-oauth-read-and-write > An OAuth connection always carries read and write.