Connect REVU to your AI assistant#

REVU runs a Model Context Protocol (MCP) server, so an AI assistant can answer questions about your organization from REVU directly: "how did my site do this week", "which pages do AI crawlers miss", "what are reviewers unhappy about this month". The assistant reads; it never changes anything in REVU.

https://mcp.revu.ai/mcp

Connect#

You need the URL and your REVU sign-in. The assistant sends you to REVU to approve the connection, then back.

AssistantHow
ClaudeAdd a custom connector with the URL, then sign in to REVU.
ChatGPTAdd a custom connector with the URL (developer mode), then sign in to REVU.
CursorUse the install link in REVU's settings, then sign in.
VS CodeUse the install link in REVU's settings, then sign in.
Claude Codeclaude mcp add --transport http revu https://mcp.revu.ai/mcp, then run /mcp, choose revu and sign in.
Any MCP clientAdd a remote server (streamable HTTP) with the URL.

When you approve, you choose which of your organizations the assistant can read. The connection acts as you: it reads what you can read in REVU, no more. You can see and disconnect your assistants in REVU at any time, and a connection ends by itself if you leave the organization or sign out everywhere.

The server speaks MCP 2026-07-28 and the earlier revisions, and keeps no session, so a connection survives REVU deploys.

What an assistant can ask#

The tools follow your organization's products. Every answer states what it covers: the period (after your plan's history limit), timezone, environment, website or app, and which traffic. Ask your assistant to quote that scope with the numbers.

ToolAnswers
account_overviewYour organization, plan (when it renews or ends, days left, and this month's allowances), products, websites and apps, what is set up (each site's SDKs, apps) with the next step for what is not, and example questions. A good first call.
behavior_overviewA site's headline numbers against the previous period: visitors, engagement, sessions, top pages, sources, countries, devices, bots and AI agents.
behavior_trendsOne event or audience over time, or by hour of day or day of week, optionally split by page, country, browser language, device, browser, source, channel, campaign tags or an event property, and compared with the period before.
behavior_funnelConversion through 2 to 8 steps, where visitors drop off, conversion per day or week of entry, and the period before.
behavior_explain_changeWhy a number moved: what changed by channel, source, campaign tags (utm_source, utm_campaign, utm_content), page, country, device and browser against the period before.
behavior_pagesEvery page with views, visitors, engagement time, scroll depth, exit rate and rage clicks, sorted the way you ask.
behavior_featuresWhich buttons, links and forms people click, and which ones they rage click, by page.
behavior_eventsWhich events a site records, with counts. Search by name and page through them.
behavior_event_propertiesWhat one event carries: its properties and their most common values.
behavior_pathsWhere visits go after a page or event, or what led to it, and where they end, optionally only for visits from one channel such as Google search.
behavior_sourcesWhere visitors come from (channels, AI answer engines, campaigns, referring sites) and which sources convert, now and in the period before.
behavior_retentionWhether visitors come back, by day, week or month, optionally split by country, device, language or an event property.
behavior_crawl_coverageWhich search and AI crawlers read your site, which pages they miss, and which are stale.
behavior_crawler_accessWhether crawlers can reach your site: what robots.txt allows each one, and, with the Server SDK, which requests your site refused or failed.
behavior_performanceCore Web Vitals (LCP, INP, CLS) as p50, p75 and p95, rated good, needs improvement or poor, by page or device.
behavior_sessionsIndividual recent visits, with where each came from, when you ask to see them, narrowed by an event, a page, a filter or the ad or campaign they came from.
behavior_saved, behavior_saved_runYour saved cohorts, insights and dashboards. Name a cohort in a question to count only those visitors, or a saved trend, funnel or retention insight to run it as you defined it.
feedback_summaryReview volume, rating, sentiment and negative share against the previous period, with ratings by version.
feedback_themesWhat reviewers talk about: topics, sentiment and intent, and what is rising.
feedback_reviewsIndividual reviews of your own apps, when you ask to read them.
reports_list, report_getYour completed REVU intelligence reports.

Assistants that support prompts also offer ready-made questions: a weekly report, what changed and why, where crawlers miss your site, review themes this month, and a product pulse that combines behavior and reviews.

Your assistant loads the tools when it connects. If a tool or option REVU added does not show up, start a new session or reconnect REVU in your assistant.

Insights first#

REVU answers with insights: counts, rates, trends, themes and findings. Individual items appear only when you ask for them, a few at a time.

  • Your own data. Individual visits and reviews of your own apps are available on request, at most 25 per answer.
  • Apps you follow but do not own, such as competitors' apps: themes and counts only, never their review text.
  • Your visitors stay private. No answer carries a visitor, person, device or IP identifier, or a review's author. Emails and phone numbers in quoted text are masked. Counts by page, country, device or source are exact, as in REVU. Under a filter on an event property or a saved cohort, 1 to 4 visitors read "fewer than 5", and a property value sent by fewer than 5 visitors is never named; nor is a campaign tag (UTM source, campaign, content or term), city or region held by fewer than 5 visitors. Average and median times, and Core Web Vitals values, are never given for fewer than 5 visitors.

REVU treats an app as yours when its store listing's developer website is a site of yours that sends REVU events, or is on one of your organization's verified domains.

Empty and missing numbers#

Every answer says what it covers and, when it is empty, limited or partial, why, so an assistant repeats the reason with the numbers.

  • An empty answer says why: the site has no data in that period (the answer names the dates it does have), the site has not sent data yet, or the question's filters matched nothing. A misspelled event, page, property value, crawler or review topic is named as unknown, with the real ones to choose from; a country, language or app the question names but REVU does not track is named as such.
  • A comparison is fair or absent. An answer compares with the previous period of the same length only when your plan can read all of it and REVU already had data when it began. Otherwise there is no comparison, and the answer says why, so a first week never reads as growth from zero.
  • Missing is not zero. On a timeline, days before your site's first event are empty, not 0. A rate with nothing to measure, such as a funnel step nobody reached, is empty too.
  • Setup gaps come with the next step. When data is missing because something is not set up, or stopped reporting, the answer says which, where its data ends, and links to where to fix it.
  • What REVU cannot see is named. Crawlers that do not run JavaScript, which includes most AI crawlers, are seen only through the Server SDK; without it, crawler answers say so and AI crawler fetch counts are left empty rather than shown as 0. Web vitals need the Web SDK's vitals plugin, and an answer without them says so.
  • History limits are said. A period that starts before your plan's history is cut to it, and the answer says so; a period entirely before it is refused with the day history starts.
  • Cut days are marked. A period that ends now says so, and on a timeline the day or hour that holds now is marked, as is the first one when the period starts partway through it, so a part of a day is never compared with a full day.
  • Too small to report says what to change. When most of a timeline's days or hours with any activity read "fewer than 5", the answer suggests a coarser step (week or total) or a longer period.

Through the read API these are each tool's note, comparison_unavailable (retention or not_tracked), scope.window.partial, a timeline point's partial and null values.

Usage#

AI assistant requests and individual items have a monthly allowance that depends on your plan. As your organization nears it, answers say so; once it is used, answers name the date it resets and how to raise it. A list cut short by the allowance says so. REVU's settings show this month's use.

Scripts and servers#

For a script or a server that cannot sign in through a browser, create a read key in REVU's settings (Professional and Enterprise plans). A read key belongs to your organization, reads one environment, and never writes.

POST https://api.revu.ai/v1/read/behavior_trends
Authorization: Bearer revu_rk_prod_...
Content-Type: application/json

{ "touchpoint": "example.com", "page": "/pricing", "period": "last_30_days" }

GET https://api.revu.ai/v1/read lists the tools your key can run, with each one's input and output JSON Schema. Every MCP tool above is available this way, with the same answers. A read key also works as a bearer token at the MCP URL, for clients configured by hand.

StatusMeaning
401No key, or the key is unknown, revoked or expired.
402This month's API requests are used.
403The plan is not active.
404No such tool for your organization.
422Invalid input, or a question the tool cannot answer as asked; the message says what to change.
429More than 60 calls a minute with this key; wait for Retry-After.