MCP Server Documentation

EdgeComet exposes your data to AI assistants through an MCP (Model Context Protocol) server. Connect it to Claude, or any MCP client, and the assistant works with your actual site data in conversation: it queries bot traffic, Evergreen Crawl, your sitemap and Search Console performance, reads pages live, researches keywords, and stages Edge SEO changes as drafts.

Endpoint: https://cloud.edgecomet.com/mcp

Authentication is OAuth: connecting sends you through a browser sign-in to your EdgeComet account, and the client only sees the websites your account can access. See Connecting a client.

What the server exposes

• The dashboard's own numbers. The same computed reports the dashboard draws, from the same cache, so a figure the assistant quotes matches the one on your screen.
• Your data, queryable. The query tool reads every bot request and the per-page SEO digest built from them, the pages, internal links, redirects and issues of an Evergreen Crawl snapshot, the URLs your XML sitemap lists, and Search Console per page, per query and per day. It filters, aggregates and joins them on the same URL. export renders the same query to a downloadable CSV when the result is too large for a conversation.
• Drills and alerts. URL-hierarchy orientation, keyword-cannibalization detail, and your alert rules with their fired alerts.
• Live page reads. Fetch any page's current content as markdown or HTML, and for your own pages, the elements matching specific CSS selectors.
• Keyword research. SERP checks, ranked keywords, keyword expansion and metrics, competitor gap analysis, and backlink summaries. These buy live third-party data on a daily per-site budget.
• Edge SEO authoring. Draft on-page changes (titles, meta tags, content edits) as rules, preview them against the live ruleset, and save them as drafts. Deployment stays in the dashboard: there is deliberately no publish tool on this surface.

The full tool list with inputs and usage notes is in the tool reference.

What you can ask

Dashboard reports answer the questions they were built for, and Data Explorer filters the columns each explorer offers. Over MCP, the assistant writes the query itself, so your question sets the shape of the answer:

• Join sources on the same URL. Bot requests, Evergreen Crawl, your sitemap and Search Console, two or three at a time, in one table.
• Group by any field. A folder derived from the URL, a bot family, a status code, or a custom extraction field such as author, category or stock status.
• Compare crawls. Put what changed on a page between two Evergreen Crawl snapshots next to the clicks it earned in each.
• Chain the steps. Run a query, read the pages it found, and stage the fix as an Edge SEO draft in the same conversation.

The questions below each ran against a live EdgeComet site. Ask them in your own words: the assistant translates them into queries. The query guide shows the queries behind several of them.

One table for every indexable page

Give me every indexable page with its number of linking pages, last Googlebot visit, last AI bot visit, clicks and impressions, as a CSV.

Combines: Evergreen Crawl, bot requests, Search Console.

The assistant joins the three sources on the URL and returns a download link from export. Sorted in a spreadsheet, the file shows pages that are linked but never fetched by Googlebot, and pages Googlebot fetches that earn no impressions.

Pages in search that answer bots with errors

Which URLs with Search Console impressions returned a 4xx or 5xx to bots in the last 30 days?

Combines: bot requests, Search Console.

Each row is a URL, the status code bots received, and how many times it happened. These are errors on pages that Google still shows to searchers.

Sitemap URLs Googlebot has not requested

Which URLs in my sitemap has Googlebot not requested in the last 30 days?

Combines: your XML sitemap, bot requests.

Useful after a launch or a migration, to see which listed pages Google has not come back for. The sitemap stores URLs in a lightly normalized form, so the assistant checks a short result against the URL text before it reports a page as missing.

Pages close to page one, and how well they are linked

Which indexable pages average position 4 to 20 with at least 20 impressions, and how many pages link to each?

Combines: Evergreen Crawl, Search Console.

A page in striking distance with few linking pages is a candidate for new internal links, which Edge SEO can insert without a code release.

What changed on pages that lost clicks

Which pages earned fewer clicks in this crawl than in the previous one, and did their title or internal links change?

Combines: two Evergreen Crawl snapshots, Search Console for each snapshot's window.

Each snapshot carries the clicks from its own window, so the drop and the change sit on one row: the old and new title, and the inlink count before and after.

Search and AI demand by section

Break down Googlebot visits, AI bot visits, AI user fetches and search clicks by top-level folder for the last 30 days.

Combines: bot requests, Search Console.

The assistant derives the folder from each URL and groups by it. AI user fetches are the real-time requests an assistant makes while answering someone, such as ChatGPT-User or Claude-User, so you see which sections AI assistants read on demand against the clicks Google sends them. See AI bots EdgeComet identifies for the bot groups.

Performance by author, category, or any extracted value

Which blog authors' articles earn the most clicks and impressions?

Combines: custom extraction, Search Console.

Any custom extraction field works as a grouping key, so the same question works for a category, a brand, or a stock status that you extract from your pages.

Pages behind AI fan-out queries

Which pages show up for long, AI-style queries without getting clicks, and how often do AI assistants fetch those pages?

Combines: Search Console queries, bot requests.

Fan-out queries are searches of eight or more words that draw impressions and no clicks, the pattern AI assistants leave when they expand a prompt into searches. The assistant finds the pages those queries land on and lists the AI bot and AI user fetches each page received.

Bots that fake their identity

Which bots are being spoofed on my site, and how many fake requests did each one make in the last 30 days?

Combines: bot requests, IP verification.

EdgeComet checks each request's IP address against the ranges the bot's operator publishes. The assistant counts the requests that claimed a bot's name from outside those ranges, per bot. Every other query leaves fake requests out by default, so they appear only when you ask about them. See Verifying that an AI bot is genuine.