Airefs MCP server tools

Every tool below is read-only: a connected assistant can read what your Airefs account can read, and it cannot change anything. That covers your profile, your sites’ analytics and prompts, and the stored AI answer texts. It also covers competitor and source lists, recommendations, and the brand context behind them.

Whatever a tool returns reaches your assistant’s provider. This page is here so you can see exactly what it reaches before you connect anything.

The tables are generated from the server’s own tool registry, so they match the list your client gets when it connects. Setup instructions live on the Airefs MCP server page.

Visibility and metrics

Clicks and impressions

Clicks measure recorded human visits referred by AI assistants. Impressions measure verified AI crawler fetches, not human views or citations. Ask for daily trends, source breakdowns, or top pages for a site and an inclusive UTC date range.

Filter by the exact assistant name or page path returned by these tools. Impressions default to user-search, matching the Impressions overview. Set intent to indexing or unknown to count that intent alone, or to all to include every verified fetch. Each call accepts up to 366 days within the site’s available history; the response reports the effective dates if history limits shorten the window.

Time series and source reports include the full filtered event total, even when their rows are truncated. Top-page lists return total_events: null; adding their rows does not give a site total. Rankings default to 25 rows; set limit to request between 1 and 200. Time series return at most 5,000 rows. All six traffic tools use window_truncated to indicate more matching rows and provide no continuation cursor. Narrow the date range or filters to investigate a truncated result.

For comparisons, request two equal-length periods with the same filters. Empty results do not confirm that tracking is installed, and dividing clicks by crawler impressions does not produce a click-through rate.

Example: top AI referral sources

First call whoami to find an accessible site_id. This example shows the parameters of an MCP tools/call request. Replace the site identifier and choose dates within your site’s available history, ending on or before today in UTC.

{
  "name": "list_click_sources",
  "arguments": {
    "site_id": "YOUR_SITE_ID",
    "start_date": "2026-10-20",
    "end_date": "2026-10-26",
    "limit": 2
  }
}

An illustrative tool result’s structuredContent could be:

{
  "columns": ["source_name", "clicks"],
  "rows": [["ChatGPT", 7], ["Claude", 6]],
  "page_row_count": 2,
  "window_truncated": true,
  "metric": "clicks",
  "definition": "Recorded human referral visits attributed to AI assistants, deduplicated by event ID. These are visits, not unique people, conversions, or citations.",
  "start_date": "2026-10-20",
  "end_date": "2026-10-26",
  "time_zone": "UTC",
  "total_events": 15,
  "coverage_note": "Only recorded events are counted. Empty results do not establish that tracking is installed or that no traffic occurred. Today may be incomplete. Never divide clicks by crawler impressions to infer a click-through rate."
}

Pair each row’s values with columns: the two displayed sources account for 13 clicks, while total_events includes all 15 clicks in the filtered window. window_truncated: true means additional sources were omitted by limit. To inspect one source, pass its exact name, such as "assistant": "ChatGPT", in another call. If history limits shorten the requested window, the result also includes history_limited with the requested and effective start dates and the history allowance.

Prompts and tags

Sources and citations

AI responses

Competitors

Articles and content

Discussions

Backlinks

Account and access

Reference and utilities

How results work

Most tools that list things answer one page at a time. The assistant asks again with a cursor to get the next page and keeps going until the server says there is nothing left. The rows in front of you are one page, not a population, so a count of them is not a count of everything.

Some totals come back marked as capped. Read those as “at least this many” rather than an exact figure. If the precise number matters to you, say so in your question, because an assistant that repeats a capped total as final is claiming more than the data supports.

Not every tool pages. Several answer with a recent window instead, which covers the question they exist for but will not reach further back when asked. Others return a single object rather than a list at all. Your client learns which is which from the server, so this mostly stays invisible until you ask for more history than a window holds.

Site names cause more confusion than anything else here. Ask which Airefs sites the assistant can see and it will name each one the way Airefs stores it, usually as the domain. Use that name in your next question and the ambiguity goes away.

Questions your plan does not cover come back as errors on purpose. Google AI Overview as an answer source and long stretches of analytics history both follow your plan, and when one of them is unavailable the tool says so rather than answering with less. That way you can tell a genuine zero from data your plan does not include.

Asking a lot of questions in quick succession can hit a rate limit. The error says how long to wait, and retrying after that usually works.