AI visibility tool with an MCP server for Claude

Connect it in Claude

• In Claude, open Customize > Connectors, click + Add, then Add custom connector. On a Team or Enterprise plan an owner adds it once under Organization settings > Connectors, and members then click Connect.
• Name it Promvia, enter https://promvia.app/api/mcp as the server URL, and click Continue.
• Under Authentication, choose to sign in now or when needed. Under OAuth client, keep Claude's published identity, the recommended option; Promvia's sign-in server also accepts automatic registration. Leave the request headers empty.
• A Promvia window opens. Sign in, check which Promvia account is being connected, and choose Allow access. Read access is included. Adding questions, starting checks, generating briefs and updating action items are separate boxes, each unticked until you tick it.
• In a chat, click + at the lower left, open Connectors, and switch Promvia on for that conversation.
• To disconnect, remove the connector in Claude, or revoke it in Promvia under Settings > Integrations > Connected apps.

Other MCP clients that can sign in with OAuth, such as Claude Code, use the same URL. A client that can only send a fixed header can use a Promvia API key as a Bearer token instead; a key reads but never writes, so a leaked key cannot change anything. The full reference is on the developers page.

What Claude can read

• list_sites: your sites and their ids. Claude calls it first, because every other tool takes a site id.
• get_visibility_score: the AI Visibility Score with its grade, your rank, your share of voice, and the five businesses with the most answer slots. Each share comes with its sample and a 95% interval, and two businesses whose intervals overlap are reported as tied, not as one ahead of the other.
• get_citations and get_evidence: per tracked question, whether the answer named you, which other domains it cited, answer excerpts, and the web searches the engine ran where it reports them.
• get_citation_history: the same checks over time, by engine, up to 90 days back.
• get_sources_report: the domains AI answers cite for your questions, ranked and grouped by kind.
• get_geo_audit, get_geo_actions and get_deep_scan: the readiness audit, the status of each fix, and page-level findings from the site scan.
• get_metrics: AI-referred visits and verified revenue by source, with unverified conversions kept separate.
• get_usage, get_opportunities and get_prompt_templates: your allowances, suggested questions to track, and question variants by country, language or persona.

Questions that work in plain language: "Which of my tracked questions does no engine name me on?", "Who is ahead of me, and is it a tie?", "Which domains do the answers cite instead of my site?", "How many visits came from ChatGPT this month?" Claude picks the tool; you can ask it which one it used.

What it can change, and only with your permission

• add_prompt and archive_prompt: start or stop tracking a question. Needs the prompts:write permission and counts against your plan's tracked-question limit.
• start_citation_run and start_deep_scan: queue a check or a site scan. Needs runs:write, and runs through the same limits, verification and budget as the button in the dashboard.
• generate_brief: write a content brief for a tracked question. Needs content:write and draws on the account's generation budget.
• update_action_status: mark a fix as open, in progress, done or dismissed. Needs actions:write.

A read-only seat on a team stays read-only through Claude, whatever the grant says. Claude can also ask for your approval before it runs a tool; its Help Center advises choosing Allow always only for servers and tools you trust.

What it is not

• Not a live question to ChatGPT. The tools read checks Promvia has already run and stored. A fresh check is start_citation_run, which queues one; the answer arrives when the check completes.
• Not a separate data set. MCP and the REST API are built on one read model, so Claude sees the same numbers, and the same caveats, as your dashboard. A score with too few answers behind it says so, and Claude has that note to quote.
• Not free of limits. REST requests and MCP tool calls share one monthly pool, stated on the developers page; checks, scans and briefs keep their own budgets.
• Not on every Promvia plan. The MCP server is on Pro and Agency. On Free and Starter, Promvia's consent screen says the plan does not include it instead of offering Allow access, and the server refuses tool calls from those plans. Paid plans have not opened yet; pricing lists what each plan includes.

The measurement behind these tools is described in how to track your brand's mentions across AI assistants, and the share that get_visibility_score reports in AI share of voice.

Frequently asked questions