MCP is live

Agent-ready SEO analytics

Connect CrawlRaven to Claude, ChatGPT, Codex, or Cursor and ask what to ship this week. Use a seven-tool free preview or up to 13 paid tools for Search Console, linked GA4, your ranked plan, and approved manual notes.

Connect your agentSee pricing
Works with any OAuth MCP client7-tool preview · up to 13 paidOAuth scopes · human approval
4.8
from 247 reviews

CrawlRaven runs a remote MCP server at mcp.crawlraven.com/mcp so AI agents can use your SEO analytics as typed tool calls. Your agent can pull Search Console performance, linked GA4 outcomes, ranked opportunities, target keywords, and the site timeline, then turn them into answers, briefs, and reports. Most tools read data. With a separate paid scope, a client may propose manual note changes, but the same authorizing user must review and approve each exact plan in CrawlRaven.

How it works

Your data, served to any agent as typed tools

The MCP server sits between your CrawlRaven data and the agents you already use. Most tools read; manual note changes require separate permission and your approval.

What the agent reads
Search performance
daily KPIs · period vs period
Linked Analytics
sessions · key events · outcomes
Queries & pages
top corpora · filterable
SEO plan
opportunities · targets · evidence
Timeline
notes · updates · approved changes
↓
MCP
CrawlRaven MCP server
mcp.crawlraven.com/mcp
OAuth · scoped · 7–13 tools
↓
Where it answers
Claude
ChatGPT
Codex
Cursor
Any MCP client
The catalog
list_websitesget_website_overviewget_search_performancelist_querieslist_pagesget_analytics_summaryget_analytics_breakdownlist_opportunitieslist_target_keywordslist_annotationsplan_annotation_changeget_annotation_change_statuslist_archived_annotationsFull reference →

What your AI agent can do with CrawlRaven

Seven bounded tools power the free preview; paid connections can expose up to 13 across approved scopes. The same report and plan data you see in the app, in a shape a model can actually use.

Ask questions in plain language

How did the site do this month? Which pages drove key events? Your agent picks the right scoped CrawlRaven tool, pulls typed Search Console or linked GA4 data, and answers in the chat.

Read the ranked plan

Opportunities arrive already ordered by estimated impact, with evidence and a recommended action attached, so your agent reasons over a decision instead of inventing one from raw rows.

Track trends, decay and causes

Daily Search Console and GA4 series, period comparisons, and a timeline of updates and notes. Enough for an agent to separate search movement from page-level outcomes and nearby events.

Draft reports in context

Have your agent turn the overview, the biggest movers, and the timeline into a monthly client update, a sprint plan, or a writer-ready checklist, inside the chat it already lives in.

Use cases

What you can do with CrawlRaven in your agent

Every CrawlRaven MCP tool is one message away. Here are the workflows teams run most — just type the prompt, and the agent picks the tools.

This week's priorities

Opportunities already carry impact estimates, evidence and a recommended action — so the agent reads a decision instead of inventing one.

What should I work on this week for example.com?
→ calls list_opportunities

The monthly client update

Overview for the headline, pages and queries for the movers, timeline for the explanation — drafted in one prompt.

Write the monthly search update for example.com: how the period compared, what moved most, and what explains it.
→ calls get_website_overview · list_pages · list_annotations

Decay triage

Find what's slipping, see which queries it lost, and check whether a Google update lines up with the drop.

Find the pages that lost the most clicks over 90 days and check whether a Google update lines up with the drop.
→ calls list_pages · list_annotations

Plan against reality

Your target keyword workspace — planned URLs, intent groups, priority — checked against what actually ranks.

Which target keywords have no page ranking yet, and what do their planned URLs rank for instead?
→ calls list_target_keywords

Query mining

Filter the top-query corpus by text and let the agent surface the impressions you're not converting to clicks.

Which queries containing pricing get impressions but almost no clicks?
→ calls list_queries

Trend forensics

Daily Search Console rows with period-over-period comparison, so the agent can find the exact week something changed.

Did traffic to example.com fall gradually or overnight? Point out the sharpest drops.
→ calls get_search_performance

Search meets outcomes

Linked GA4 totals and bounded breakdowns add sessions and key events without pretending a Search Console query caused a conversion.

Which landing pages combined search growth with meaningful sessions and key events over the last 28 complete days?
→ calls list_pages · get_analytics_breakdown

Keep the timeline current

The agent can propose a manual note, but planning changes nothing until the same authorizing user reviews and approves it in CrawlRaven.

Plan a note for the homepage launch on August 30. Show me the exact proposal and wait for my approval.
→ calls plan_annotation_change
A full run

From question to evidence-backed answer

This is the decay-triage workflow from the docs, end to end: the agent picks the tools, reads typed rows, lines the drop up against the timeline, and separates a site problem from a Google one.

No export, no screenshot — the agent reads the actual rows
The timeline keeps it honest: updates checked before theories
Follow-ups are one message, not another afternoon
Steal the workflow prompts →
your AI assistant
Find the pages that lost the most clicks over 90 days and check whether a Google update lines up with the drop.
⚙list_pages · days: 90
⚙list_annotations · days: 90
Three pages account for most of the decline:
/blog/seo-audit-checklist−412 clicksdrop starts wk of Jun 28
/tools/meta-tag-checker−188 clicksdrop starts wk of Jun 28
/blog/gsc-guide−121 clicksgradual since early Jun
The June 28 core update on your timeline sits inside the drop window for the first two. The third began slipping three weeks earlier — that one is on the site, not on Google.
Turn the first two into refresh briefs I can hand to a writer.
drafting…

Illustrative transcript — the prompts are real, from the MCP docs.

Connect with OAuth

Connect CrawlRaven to your agent

Point any OAuth-capable MCP client at the remote server. MCP is the shipped external integration; CrawlRaven v2 does not expose a supported customer REST API.

Full setup steps, the tool list, scopes, limits, and a prompt library live in the MCP server documentation.

OAuth · no manual token required

Add the remote MCP URL in an OAuth-capable connector. The connector discovers CrawlRaven's auth metadata and prompts you to sign in before sending tool requests.

Connect Claude

  1. 1Open Claude settings and go to Connectors.
  2. 2Add a custom connector named CrawlRaven.
  3. 3Paste the endpoint URL below and complete the CrawlRaven sign-in prompt.
Endpoint URL
https://mcp.crawlraven.com/mcp
Advanced OAuth metadata URLs

Most users only need the endpoint URL above. These are for clients that inspect OAuth discovery directly.

Protected resource metadata
https://mcp.crawlraven.com/.well-known/oauth-protected-resource
Authorization server metadata
https://app.crawlraven.com/.well-known/oauth-authorization-server

Why ask an agent instead of opening a dashboard

A dashboard is good at showing you everything. An agent is good at answering one question and then acting on the answer. Most SEO work is the second kind: what changed, why, and what should we do about it. Which is why the exports and copy-paste steps in a normal reporting workflow take longer than the analysis.

The agent reads structured data, not a screenshot

This is the part people underestimate. Pasting a screenshot of Search Console into a chat model gives it a picture of a table. Over MCP, the model receives the actual data: daily performance rows, query and page tables, opportunities that already carry impact estimates, evidence, and a recommended action. So it filters, sorts, and reasons instead of squinting. MCP is the supported external integration; CrawlRaven v2 does not expose equivalent customer REST endpoints.

  • No export step. The numbers arrive in the conversation, so there is no CSV round trip between spotting a drop and explaining it.
  • Follow-up questions stay cheap. Asking which of these pages actually get traffic is one more message, not a second pass through a spreadsheet.
  • The output is deliverables, not just charts. The same context that found the biggest movers can write the monthly client update that explains them.

CrawlRaven is not the only SEO server an agent can call, and not the widest. It is the one serving your own Search Console and GA4 rather than a rented index. We compared it against eight others, and the plan each vendor requires, in our roundup of the best SEO MCP servers. And because Google never shipped an official Search Console MCP server, we also wrote up every real route to getting GSC data in front of an agent, including the two that are not ours.

What to hand the agent, and what to keep

The division that works is simple: the agent reads, sorts, drafts, and may propose a manual timeline note; you decide and verify. Most MCP tools cannot change anything. A note proposal still changes nothing until the same authorizing user reviews and approves its exact values in CrawlRaven.

  • Good agent work: summarising a period, comparing Search Console with linked GA4 outcomes, finding the sharpest drops, turning opportunities into a checklist, drafting the client update, and proposing a concise manual note for review.
  • Keep for yourself: deciding what actually gets written or fixed, judging business impact, approving any manual note mutation, and signing off anything that goes to a client.
  • Always verify: check the numbers in anything client-facing against the app before you hit send.

The failure mode to plan for

A model asked to explain a traffic drop will always produce an explanation. That is why the timeline tool exists: it lets the agent check the boring causes first, a Google update, an import, a note you left yourself, before it invents a clever one. Prompt for evidence, not just narrative, and the workflow holds up.

Getting from zero to an agent that knows your site

The setup is one-time. After that, an answer is a sentence away. For the per-client version of these steps, and the four failures that account for most bad connections, see how to connect Claude to an SEO MCP server.

  1. 01

    Start with CrawlRaven access

    Eligible free accounts can try seven bounded read-only tools for one pinned site. Paid access unlocks broader periods, pagination, linked GA4 tools, active annotations, and optional approved manual note management.

  2. 02

    Connect a website and Search Console

    The core tools serve the same Search Console and plan data as the app. Link GA4 to the website as well if you want paid Analytics summary and breakdown tools.

  3. 03

    Add the MCP server to your client

    Add mcp.crawlraven.com/mcp to Claude, ChatGPT, Codex, Cursor, or any OAuth-capable MCP client. Browser OAuth shows each requested scope; reconnect an older grant to add View analytics or Manage annotations.

  4. 04

    Ask for your websites

    list_websites is the natural first call: it returns the website_id every other tool needs, so "which sites do I have in CrawlRaven?" gets the agent oriented.

  5. 05

    Ask questions, not for dashboards

    "What should I work on this week?" "Did traffic fall gradually or overnight?" "Which queries containing pricing get impressions but no clicks?" The agent picks the tools.

  6. 06

    Verify anything client-facing, then make it a habit

    Spot-check the numbers against the app before a report goes out. Once you trust the loop, the monthly update becomes a single prompt.

Works with any MCP-compatible agent

MCP is an open standard, so CrawlRaven plugs into the agents you already use, and any new one that adopts the spec.

Claude
ChatGPT
Claude Code
Codex
Cursor
OpenClaw

…and any other MCP client that supports remote servers with OAuth.

AI SEO agent & MCP FAQ

An AI SEO agent is an assistant such as Claude, ChatGPT, Cursor, or another MCP-compatible client that can call tools. With CrawlRaven connected, it can read scoped Search Console reports, linked GA4 outcomes, opportunities, target keywords, and timeline data, then turn them into answers and reports. With separate permission it may also propose a manual note for your approval.
Free plan — no credit card

Hand your agent the plan

Grab a lifetime license, connect the MCP server, and ask your agent what to ship this week.

Get lifetime accessRead the MCP docs
3
Data sources joined
200+
Point audit checks
1
Ranked plan out