SEO MCP server: your own search data inside Claude and Cursor

What the MrKeyword MCP server is

The Model Context Protocol (MCP) is an open standard that lets an AI assistant call tools on another service. MrKeyword runs a hosted MCP server, so Claude or Cursor can look things up in your MrKeyword workspace while you talk to it: how the site is doing, which queries slipped, what the last crawl found, and which page deserves attention first.

The difference from pasting a CSV into a chat is that the assistant asks for exactly the rows it needs, when it needs them, and every row carries its source and the date it was observed. You stay in the conversation; the data stays current.

Connect it in two minutes

1. Create a token

   In MrKeyword, open Settings → Connections → API & MCP Tokens and create a token. Leave “Allow this token to make changes” unticked for a read-only token. The token is shown once; store it like a password.

2. Add the server to your assistant

   For Claude Code, one command does it:

   claude mcp add --transport http mrkeyword https://app.mrkeyword.io/mcp \
     --header "Authorization: Bearer mk_your_token_here"

   For Cursor, add it to .cursor/mcp.json in a project, or ~/.cursor/mcp.json for every project:

   {
     "mcpServers": {
       "mrkeyword": {
         "url": "https://app.mrkeyword.io/mcp",
         "headers": { "Authorization": "Bearer mk_your_token_here" }
       }
     }
   }

   In the Claude desktop app, go to Settings → Connectors → Add custom connector, enter the URL above and add an authorization request header with the value Bearer mk_….

3. Ask a question

   Try “How did my site do this week?” or run the weekly_seo_review prompt. The developer reference has the same setup for plain HTTP and every tool's arguments.

What your assistant can read

Four ready-made prompts

Prompts are saved workflows the server offers. In Claude they appear as slash commands, so the common jobs are one click away:

• weekly_seo_review: rank summary, movers and what needs attention, ending in five actions for the week.
• fix_my_top_issues: the latest site audit, grouped by category, with concrete fix steps ordered by impact.
• find_keyword_opportunities: near-miss queries at positions 11–20 turned into one prioritised list. It tells you the lookup cost before it sizes anything.
• client_report: rankings, queries and the audit written up as a client-ready report.

Questions worth asking

• “Which keywords dropped this month, and is the right page ranking for them?” — get_rankings with the moved_down and wrong_page filters.
• “Which pages could use a little love?” — get_search_queries (almost there) plus get_audit_issues for those pages.
• “Write a brief to refresh my pricing page.” — get_page_brief, built only from that page's queries and audit.
• “Where should I add internal links this week?” — get_internal_link_suggestions.
• “Does ChatGPT mention us yet?” — get_ai_visibility, from your tracked questions.
• “Draft this month's report for the client.” — the client_report prompt.

Built so you stay in control

• Read-only unless you say otherwise. The action tools refuse a read-only token with a plain message telling you how to make one that allows changes.
• One workspace per token. A token can never touch billing, team members, invitations or other tokens. Workspace viewers can only create read-only tokens.
• Logged and revocable. Every action is written to the workspace audit log against the token and the person who made it. Revoke a token from the same screen in one click.
• Honest numbers. Positions are Search Console's average position for a period, not a live rank check. Estimated volumes carry their label. Every payload is dated with as_of.

What the MCP server does not do

• It does not return backlinks, competitor rankings or traffic forecasts. No tool reads them, so no answer should contain them. Use the Data API for domain-level data.
• It cannot change your website, your Search Console property or your Google account. Actions only change things inside MrKeyword.
• It does not stream or push. The endpoint answers one JSON-RPC request per POST, up to 120 calls a minute per token.
• It does not run a model for you. Your assistant does the writing, on your assistant's plan.

Questions and answers