CrawlRaven MCP

CrawlRaven MCP is a remote Model Context Protocol server that plugs into Claude, ChatGPT, Cursor, Claude Code or any OAuth-capable MCP client and lets you run your whole SEO operation from the chat. It exposes the data CrawlRaven already holds — Search Console, linked GA4, ranked opportunities, your target keyword plan and your site timeline — as thirteen typed tools, so your agent reads the actual rows and tells you what to fix, refresh and write next. It is built for anyone who decides what gets written, refreshed or fixed next and would rather ask than export: SEO agencies, in-house SEO and content teams, and founders or solo marketers. Connection is one URL and an OAuth sign-in, and twelve of the thirteen tools are read-only by default. The problem CrawlRaven MCP addresses is the gap between the questions you actually have and the way Search Console answers them. Search Console answers one filtered question at a time: the Performance report caps every table at 1,000 rows, so the long, specific queries that AI search sends fall off the end, and every other angle means another filter and another export. Matching queries to pages, then pages to GA4, is manual work, so it happens once a quarter — by which time the decaying page has already lost its clicks. And when a raw CSV is pasted into a chat, the model cannot tell whether a drop lines up with a Google update or which page should own a query, so it fills the gap with a plausible story. CrawlRaven MCP replaces export, filter, match, repeat with one question and a ranked list. Four of the thirteen tools read Google Search Console. The agent can search queries by text instead of scrolling a table, narrow any report to one exact page in a single call, and pick any window from 1 to 365 days. Overview, series, queries and pages are all exposed, so the model works on the site's real performance data rather than a screenshot of a report. Because the tools return typed rows with the period and any truncation stated, answers can name the tool and the date range they used — and long, specific queries that fall off the end of a 1,000-row Performance report stay reachable. Two more tools join GA4 outcomes, returning sessions, engagement and key events for the linked property so Search Console and linked GA4 answer in the same conversation. One tool ranks what to fix next: opportunities arrive ranked with the evidence attached and an action to take, covering striking distance rankings, CTR gaps and content decay. Another checks your keyword plan, telling the model which page is meant to rank for a query and surfacing target keywords with no page ranking yet or query conflicts where no page owns the term. A further tool explains drops using your timeline, which shows verified Google updates next to your own notes. Finally, three tools write notes you approve — including one that plans a timeline note and waits for your approval in the app. Scopes and defaults shape what a connection can do. Each tool belongs to a scope you approve in the browser, and a connection only sees the tools you granted. Twelve tools only read; the single tool that plans a timeline note waits for your approval in the app. There are no API keys and no manual tokens to paste. This means a connected client cannot quietly change anything in your account, while the agent still has enough context — sites, rankings, analytics outcomes, opportunities, keyword plan and timeline — to give an answer you can act on. The server sits between your CrawlRaven data and the AI client you already use. The client picks a tool, CrawlRaven checks the scope, and typed rows come back with the period and any truncation stated. Setup is one-time and takes about two minutes: connect Search Console in CrawlRaven, and link GA4 as well if you want the analytics tools; paste mcp.crawlraven.com/mcp into your client using the remote HTTP transport; sign in and approve the scopes in a browser; then ask your first question, starting with which sites you have in CrawlRaven. The connector discovers CrawlRaven's auth metadata and prompts you to sign in before sending tool requests, and advanced OAuth metadata URLs are published for clients that inspect OAuth discovery directly. The documented workflow runs end to end without an export or a screenshot: the agent lists pages that lost the most clicks over 90 days, reads the annotations timeline, lines the drop up against a Google update, and separates a site problem from a Google one. Outcomes reported by customers include a daily and Monday content refresh routine where Claude and CrawlRaven update around five to seven pages every Monday, taking a daily SEO routine from two hours to five minutes and generative AI overview impressions from roughly 2,500 to about 9,000 a day. Another customer cut identifying the next set of keywords from three hours to thirty minutes and now describes in plain English what data it needs and gets all of it in seconds. A third moved fixing low-hanging fruit from quarterly to weekly and checks every new post against what the site already ranks for, so pages do not cannibalise each other. Underneath these numbers is a simpler benefit: follow-ups become one message rather than another afternoon of exporting. Concrete workflows where the server is used include weekly content refresh at scale, next-set keyword discovery, pre-publish cannibalisation checks, and decay triage that starts with a question such as which pages lost clicks this month or did traffic fall gradually or overnight. Users also ask which queries containing a term get impressions but no clicks, find pages sitting between position 8 and 20, group every query for a URL by intent, ask which landing pages drove key events over the last 28 days, turn the top five opportunities into a checklist for a writer, and plan a note for a launch that waits for approval. Weekly and monthly reporting can be written from live data, and answers cite their source so they hold up on a client call. Supported clients are Claude and Claude Desktop, Claude Code, ChatGPT, Codex, Cursor and any OAuth-capable MCP client, connected over the remote HTTP transport at https://mcp.crawlraven.com/mcp. Pricing starts with a free preview at $0 for eligible free accounts: seven read-only tools, one pinned website, a fixed 28-day window, up to 50 rows per call and 20 tool calls per account per day. Full access is included with every lifetime license as a one-time payment with no MCP add-on, and unlocks all thirteen tools by the scopes you approve, every website on your plan, custom ranges from 1 to 365 days, and the linked GA4 and timeline tools. In short, CrawlRaven MCP turns Search Console, GA4, your keyword plan, your ranked opportunities and your timeline into tools your AI agent can call, so you ask what to ship this week instead of exporting what shipped last quarter.