Airefs

Ask Claude how your mention rate moved last month, which competitors showed up in answers where your brand did not, which sources keep getting cited in your category, and which discussions are worth joining this week. The answers come from your own Airefs account rather than from the model’s general memory of the internet, and you get them without opening the dashboard or exporting anything.

The Airefs MCP server is what connects the two. The Model Context Protocol (MCP) is a shared standard for plugging AI apps into outside services, in the way USB-C is one shape of plug that fits many devices. You add one URL to your assistant, sign in once, and it can read your Airefs data from then on.

What it covers

Your assistant can read mention rate, citation rate, and average mention position over time, broken out by prompt, by competitor, and by answer source. The whole Sources layer comes with it: the domains and URLs collecting citations, and the exact AI responses behind each one. If a number is on your overview, a tool can reach it.

The opportunities to move those numbers are covered too. Article recommendations arrive with the backlog and the pipeline they feed, discussions covers the threads AI answers are already citing, and backlinks is the outreach list plus the citation evidence behind each domain.

Two things the server does not expose: Impressions, the visits AI crawlers make to your pages, and Clicks, the visitors who arrive from an AI answer. Both live in the dashboard.

Connect

The server URL is https://mcp.getairefs.com/mcp. That is the only thing you need to copy.

Add it in your client, then sign in with your Airefs account in the browser tab that opens. There is no API key to copy anywhere: the sign-in hands your client short-lived tokens, which it stores and you can revoke. After you sign in, the connection reads what your account can read: your sites’ data plus your profile basics. Nothing more.

Works the same in Claude on the web and in Claude Desktop.

1. Open Settings → Connectors.
2. Click +, then Add custom connector.
3. Name it Airefs and paste https://mcp.getairefs.com/mcp as the remote MCP server URL.
4. Click Add, then Connect.
5. Sign in with your Airefs account in the window that opens and approve access.
6. Start a chat and mention Airefs in your question. If the tools do not appear, open the tools menu in the chat box and switch the Airefs connector on.

Custom connectors work on every Claude plan, but a free account can keep only one at a time, so remove another before adding Airefs. On Team and Enterprise plans only an owner can add the connector for the organization, under Organization settings → Connectors; members then connect to it individually.

Custom MCP servers live behind ChatGPT’s developer mode, which runs in the web app only and covers Pro, Business, Enterprise, and Edu. Plus does not support custom MCP servers. Pro accounts are limited to servers with read and fetch tools, which is all Airefs has.

On Pro you add it yourself:

1. Open Settings → Apps → Advanced settings and turn Developer mode on.
2. Go back to Apps and click Create.
3. Name it Airefs, paste https://mcp.getairefs.com/mcp including the /mcp path, and choose OAuth as the authentication method.
4. Click Scan Tools, sign in with your Airefs account in the window that opens, then click Create.
5. Start a new chat and pick Airefs from the tools menu.

On Business only admins and owners can use developer mode at all, and each one turns it on for their own account. An admin creates the app from Workspace settings → Apps → Create, then publishes it from Drafts; ordinary members have no developer mode toggle and get Airefs once it is published. On Enterprise and Edu an admin grants developer access first, under Permissions & Roles → Connected Data, after which authorized members turn developer mode on in their own settings, but only admins and owners can publish for the workspace.

Register the server once from your terminal.

claude mcp add --transport http airefs https://mcp.getairefs.com/mcp

1. Run /mcp inside Claude Code and choose Authenticate for airefs. A browser window opens for sign-in.
2. Run claude mcp list to confirm the server is connected.

To share the server with everyone working in a repository, add the scope flag between the server name and the URL:

claude mcp add --transport http airefs --scope project https://mcp.getairefs.com/mcp

Each person still signs in with their own Airefs account.

Add the server to .cursor/mcp.json in your project, or to ~/.cursor/mcp.json to use it everywhere.

{
  "mcpServers": {
    "airefs": {
      "url": "https://mcp.getairefs.com/mcp"
    }
  }
}

1. Save the file, then open Cursor Settings → MCP.
2. Click Login next to the Airefs server and sign in with your Airefs account.
3. Ask about your Airefs data in the chat pane once the server shows as connected.

Add the server to .vscode/mcp.json in your workspace.

{
  "servers": {
    "airefs": {
      "type": "http",
      "url": "https://mcp.getairefs.com/mcp"
    }
  }
}

1. Save the file and click Start above the server entry.
2. Sign in with your Airefs account in the browser window that opens.
3. Open Copilot Chat in agent mode and pick the Airefs tools from the tools picker.

Commit the file so the whole team picks up the server. Sign-in stays per person, so nobody inherits anyone else’s access.

Any client that speaks streamable HTTP and supports OAuth can use the server directly. Give it the URL https://mcp.getairefs.com/mcp, then complete the browser sign-in when the client asks for it. There is nothing else to configure.

For clients that only run local servers, bridge the connection with mcp-remote, which needs Node.js installed:

{
  "mcpServers": {
    "airefs": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.getairefs.com/mcp"]
    }
  }
}

On its first run mcp-remote opens a browser window for sign-in and remembers the result, so later runs start without a prompt.

What your assistant can see and do

Every tool on the Airefs MCP server is read-only, without exception. An assistant can read your Airefs data and it can never change anything in your account. It cannot add or remove prompts, dismiss an opportunity, schedule an article, edit your competitor list, or touch billing. So the worst a bad question can do is read something you would rather not have shared, because everything a tool returns goes to your assistant’s provider and is then covered by that provider’s terms.

Sign-in happens through OAuth in your browser, using the Airefs account you already have. You never handle a long-lived key. The client stores the tokens it gets back, but they are short-lived and you can revoke them from Airefs whenever you want.

What the connection can reach is what your account can reach. That starts with your profile basics: name, email address, and organization memberships. It covers your sites’ visibility analytics and prompts, and the stored AI answer texts behind them.

It also includes competitor and source lists and article and backlink recommendations. Any brand, audience, or writing context you have configured is readable too.

If you belong to an organization, that means the organization’s sites; if you have one site, that means one site. Connecting a client does not widen your own access, and it does not expose other people’s sites to you. The privacy policy says the same thing in its own words.

To cut a client off, revoke it in Airefs. Open Settings → Account, find the Airefs MCP server card, and click Disconnect next to the client under Connected clients. That deletes its tokens, so it cannot ask for new ones, and the server stops answering it within a minute.

Then tidy up on the client side so the dead entry stops showing:

• Claude: Settings → Connectors, then remove the Airefs connector.
• Claude Code: claude mcp remove airefs.
• Cursor and VS Code: delete the airefs entry from mcp.json.

Doing only this second step is not enough on its own. It stops that copy of that client from asking, but the access it was granted is still live until you revoke it.

Plans and limits

MCP access is included on trial, so you can connect a client before you pay for anything. Trial accounts have tighter rate limits than paid accounts, which matters mainly for long analyses that page through a lot of rows.

Some data follows your plan the same way it does in the app. Google AI Overview is available as an answer source when your site has it enabled, and how far back analytics history goes depends on your plan. When a question falls outside what your plan covers, the tool returns a clear error saying so instead of quietly answering with less.

Recipes

The server ships six ready-made prompts. In Claude they sit in the plus menu next to the message box once the Airefs connector is on. Pick one and the assistant runs a whole procedure against your data instead of a single lookup.

Both fields a prompt asks for are optional. Leave the site field empty if your account has one site, or type the domain to choose. The focus field narrows the run in your own words: “last 14 days”, “Google AI Overview only”, “just the pricing tag”. Without a focus, prompts use ChatGPT as the answer source and a recent window.

Clients that do not surface MCP prompts can still ask for the same analysis. Every prompt below ends with a plain-language version to paste into a chat.

Weekly pulse

airefs_weekly_pulse compares this week with the one before it: how your mention rate and citation rate moved, where you stand against tracked competitors right now, which domains are collecting the most citations, and which article recommendations are active. It reads as a Monday-morning digest you can paste into a team channel. When a week has no completed data it says so rather than reporting a zero.

“Use Airefs to give me a weekly pulse for acme.com: how did my mention rate and citation rate move against last week, which source domains are cited most, and which article recommendations are active?”

Competitor radar

airefs_competitor_radar checks every active tracked competitor for real movement in mention rate or citation rate between two back-to-back periods, then separates the ones that moved from the ones that held steady and puts the current leaderboard underneath. Run it monthly, or in the week after you notice a competitor shipping something. Movement comes back in percentage points, and it never claims to know the cause.

“Use Airefs to compare each tracked competitor for acme.com this month against last month, and flag anyone whose mention rate or citation rate moved by five points or more.”

Visibility-drop investigation

airefs_visibility_drop_investigation takes a decline apart: the rate change itself, the prompts that deteriorated most with the responses behind them, which competitors moved the other way, and how the cited source domains shifted. Explanations come back sorted into supported, possible, and not testable, so you can tell evidence from a hunch. This is the one to reach for on the day someone asks why the number went down.

“Use Airefs to investigate why acme.com’s mention rate fell over the last two weeks: which prompts lost mentions, which competitors gained, and what changed in the cited sources?”

Source authority review

airefs_source_authority_review maps the sources behind your answers. It groups the cited domains and pages into your own, your tracked competitors’, and everyone else’s, then shows which domains are gaining citations and which appeared for the first time. Authority here means how often and how widely something is cited; Airefs has no third-party quality score and the prompt will not invent one. Worth running before you decide where to pitch a guest post or which publications to chase.

“Use Airefs to review the source domains cited for acme.com, split them into mine, competitors’, and third-party, and show which ones are gaining citations.”

Visibility gaps

airefs_visibility_gaps enumera los prompts donde tu marca no se menciona o se menciona raramente en un tramo reciente de ejecuciones, agrupados por país y etiqueta. Tu contexto de marca hace la clasificación: los prompts que están fuera de lo que vendes caen en fallos esperados, todo lo demás cae en la lista que vale la pena corregir, y las recomendaciones de artículos activas se emparejan con las brechas que abordan. Ejecútalo cuando estés planificando el próximo trimestre de contenido.

“Usa Airefs para listar los prompts donde acme.com no se menciona o se menciona raramente, agruparlos por etiqueta y decirme qué recomendaciones de artículos activas los abordan.”

Plan de acción

airefs_action_plan es el prompt de “qué hago al respecto”. Extrae la visibilidad actual, los prompts con brechas, las recomendaciones de artículos frente al pipeline de contenido, las discusiones citadas junto con los hilos monitoreados de Reddit y las oportunidades de backlinks en una lista priorizada construida alrededor de tres movimientos: publicar los artículos respaldados por recomendaciones, unirte a las discusiones que ya se están citando y ganar citas en los dominios en los que esas respuestas ya confían. Comienza un sprint con él y trabaja la lista hacia abajo.

“Usa Airefs para construir un plan de acción para acme.com: ¿qué debería publicar, a qué discusiones debería unirme y qué dominios debería buscar para backlinks?”

Preguntas más largas para pegar

Los prompts anteriores cubren los trabajos habituales. Cuando quieras dirigir el análisis tú mismo, o tu cliente no tenga ningún menú de prompts, pega uno de estos y cambia tu propio dominio.

Investigar una caída de visibilidad

Una caída merece evidencia antes de que tenga una historia, así que esto pide la descomposición primero.

Use Airefs for acme.com. Our mention rate is down and I have to explain it
tomorrow morning.

Compare the last 14 days with the 14 days before that on ChatGPT. Show me
the prompts that lost the most brand mentions, whether any tracked competitor
gained over the same period, and whether the domains cited for those prompts
changed.

Be clear about what the data supports and what it doesn't.

Encontrar discusiones que valga la pena unirse

Las discusiones que ya aparecen como fuentes citadas son las que las respuestas de IA están leyendo. Preguntar por el prompt citado junto con cada URL te dice qué escribir realmente en el hilo.

Use Airefs for acme.com. I have about an hour this week to reply in places
that AI answers actually cite.

List the discussion URLs that already show up as cited sources, and separately
the Reddit threads Airefs is monitoring. Drop anything where we're already
mentioned or that's marked done. For each one left, tell me which prompt it
was cited for so I know what angle to write from.

Construir una lista corta de outreach de backlinks

Las oportunidades de backlinks son los dominios de terceros que tus respuestas ya citan, con tu propio sitio y los competidores rastreados filtrados. Extraer sus páginas citadas en la misma pregunta te ahorra una pasada de investigación antes de escribir.

Use Airefs for acme.com. I'm building an outreach list for next month.

Show me the backlink opportunities ranked by how often those domains are
cited, and leave out anything already marked emailed or dismissed. For the
top ten, list the pages of theirs that get cited so I can see what they
publish about before I write the email.

Preparar un informe de competidores

Útil la semana antes de una revisión de marketing. La última línea importa: sin ella, los asistentes tienden a narrar razones que los datos no pueden respaldar.

Use Airefs for acme.com. I need a one-page competitor read for our monthly
marketing review.

For each tracked competitor: current mention rate and citation rate, the
change since the previous month, and where we sit in that ranking. Then show
which of their pages are cited in the answers we track.

Keep it factual. No theories about why anything moved.

Primeras preguntas para probar

Tres cortas, para cuando acabas de conectar a un cliente y quieres ver algo que regrese:

• “Usa Airefs para resumir mi visibilidad actual de IA y compararla con competidores activos.”
• “¿Qué prompts mencionan raramente mi marca y qué recomendaciones de artículos debería publicar?”
• “Muéstrame las discusiones citadas y los dominios de backlinks donde mi marca tiene la oportunidad más clara.”

Solución de problemas

Mi asistente no puede encontrar mi sitio

Pídele que liste los sitios de Airefs a los que tienes acceso. Nombrará cada uno exactamente como Airefs lo almacena, que suele ser el dominio. Usa ese nombre, o el dominio por sí solo, en tu próxima pregunta. Si tu sitio no está en la lista, la cuenta de Airefs con la que iniciaste sesión no tiene acceso a él. Verifica que iniciaste sesión con la cuenta correcta y pide a un administrador de tu equipo que te agregue al sitio si perteneces a una organización.

Una pregunta sobre datos de Google AI Overview devolvió un error

Google AI Overview es una fuente de respuestas separada y está habilitada por sitio. Cuando está desactivada, las herramientas lo dicen en lugar de mezclar números de ChatGPT. Haz la misma pregunta sobre ChatGPT o consulta Planes y precios para habilitar la fuente de respuestas.

Una solicitud de un tramo largo de historial puede fallar de la misma manera, porque la profundidad del historial analítico sigue tu plan. Pide una ventana más corta y la misma pregunta funciona.

La conexión expiró y necesito iniciar sesión de nuevo

El inicio de sesión no dura para siempre. Cuando tu asistente informe que no está autorizado, reconéctate desde el cliente: en Claude, abre Configuración → Conectores y haz clic en Conectar en el conector de Airefs; en Claude Code, ejecuta /mcp y autentícate de nuevo; en Cursor y VS Code, usa el aviso de inicio de sesión en la entrada del servidor. No es necesario restablecer nada en el lado de Airefs.

Mi asistente dice que está limitado por tasa

El servidor limita la rapidez con la que una cuenta puede llamarlo. El error lleva un tiempo de espera sugerido, y reintentar después de esa espera suele funcionar. Las preguntas amplias alcanzan el límite antes porque recorren más datos, así que reduce la pregunta: un sitio, una fuente de respuestas, un rango de fechas más corto. Las cuentas de prueba tienen límites más estrictos que las de pago.

No puedo agregar el conector en Claude

En el plan gratuito de Claude puedes mantener un conector personalizado a la vez, así que elimina el que tengas antes de agregar Airefs. En los planes Team y Enterprise, un propietario o propietario principal tiene que agregar el conector para la organización primero; hasta entonces, los miembros no lo verán como una opción. Claude se conecta desde la nube de Anthropic, por lo que el servidor tiene que ser accesible a través de internet público, que es lo que https://mcp.getairefs.com/mcp es.

El asistente responde sin usar Airefs

La mayoría de los clientes solo recurren a una herramienta cuando la pregunta lo hace obvio. Di “usa Airefs” en la pregunta, o nombra lo que quieres por su término de Airefs: tasa de mención, prompts, fuentes de respuestas, competidores, discusiones. Revisa también el menú de herramientas en tu cliente, ya que algunos clientes dejan un servidor recién agregado desactivado en chats existentes.

Los números no coinciden con el panel

Verifica primero la ventana de fechas y la fuente de respuestas. Una ventana diferente o una fuente diferente legítimamente dan un número diferente, y ambos lados pueden tener razón. También vale la pena verificar la granularidad: get_visibility_timeseries a granularidad semanal cuenta un prompt como mencionado si alguna ejecución de esa semana te mencionó, lo cual no es la misma aritmética que una tasa por lote.

Si todo eso coincide y la brecha sigue ahí, pregúntale al asistente qué herramienta llamó y qué parámetros usó. Un asistente que no puede decirlo está adivinando en lugar de leyendo.

¿Quieres saber exactamente a qué puede acceder un asistente antes de conectarte? La referencia de herramientas lista cada herramienta en el servidor, qué devuelve y sus parámetros clave.