Start

ScopeMCP documentation

ScopeMCP is a read-only search-data connector for AI agents. Six sources, thirty-seven tools, no writes.

Read-only by design

ScopeMCP reads Google Search Console, Bing Webmaster Tools, Analytics 4, PageSpeed Insights, Chrome UX Report and live Google result pages. Its tools do not modify sites, submit sitemaps, request indexing, or change source settings. There is no write path in the tool handlers at all — this is enforced by the code, not by a policy page.

Every response also reports what the data cannot support. If Search Console returns no coverage state for a URL, the tool says so rather than inferring one. If CrUX has no field data for a page, the tool says that is not the same as good performance. If a keyword could not be checked on a live result page, it is returned as unverified instead of as a confirmed gap.

How it works

Six steps, and the first three take about a minute each.

  • Sign in. Authorise the read-only Google permission once. You can also bring your own OAuth app.
  • Connect your sources. Search Console is part of onboarding. Analytics 4 uses the same consent screen. Bing and the competitor engine are optional and can be added later.
  • Connect an agent. Add the remote MCP endpoint and complete the ScopeMCP authorisation.
  • Ask. The agent picks a tool and reads the data from the right source.
  • Set a monitor. Optional, but this is what replaces re-asking next week.
  • Share. Any report can be exported or turned into a revocable public link.

Quick start

Add this remote MCP endpoint to your agent:

bash
https://scopemcp.galavic.world/mcp

Browser-based OAuth

ScopeMCP uses browser-based OAuth. Do not paste Google tokens, refresh tokens or API keys into an agent.

Capabilities

What the tools can and cannot answer.

  • All properties — every Search Console, Bing or Analytics property the connected accounts can reach.
  • Search performance — clicks, impressions, CTR and average position by query, page, country, device, date or search appearance.
  • Performance comparison — two explicit windows, ranked by what actually changed.
  • Search opportunities — striking distance, high-impression low-CTR rows, page-2 queries, declines and possible query overlap.
  • URL inspection — index status, canonical selection, crawl details, mobile usability, AMP and rich results.
  • Sitemaps — submitted status, warnings, errors and indexed counts.
  • Bing crawl health — crawl, index, HTTP status groups, robots exclusions and malware findings.
  • Behavioural outcome — Analytics 4 sessions, key events and revenue, including a bridge that puts them beside Search Console clicks for the same pages.
  • Page performance — Lighthouse lab audits and Chrome UX Report field data, kept clearly separate.
  • Competitor position — live result pages, measured search counts, keyword gaps, share of voice and rival movement.
  • Monitoring — six scheduled checks with in-app, email and webhook delivery.

Example prompts

Prompts that work because the tools line up with the question.

text
Show me every Search Console property I can access.

Compare clicks and impressions for the last 28 days with the previous 28.
Show the largest declines by page.

My /pricing page lost traffic. Was it ranking or click-through?

Give me the SEO health score and explain every component, including
which parts of the model could not run.

What is competitor.com ranking for that I am not? Which of those
keywords have more than 100 searches a month?

Find high-impression low-CTR queries. Inspect the strongest matching
page before you recommend changes.

Watch my top ten pages weekly and email me if any of them stops being
indexed.

Product boundaries

Stated plainly so an agent does not have to guess.

  • No tool reads page content. Nothing here can judge your copy, your schema markup or your internal links.
  • No tool crawls your site, and no tool estimates your traffic or your revenue.
  • Competitor data comes from public result pages. It shows who ranks, never who earns.
  • Benchmark CTR comes from published click-through studies. It is an industry average used for prioritisation, not a target.
  • Topic clusters are lexical groupings of observed queries, not a semantic audit and not an authority score.
  • A health score component that cannot be measured is excluded from the score and listed in gaps. It is never scored as zero.