MiranteSEO

• Three sources, one dashboard

  Search Console by query, page and period, GA4 with channels, engagement and conversions, and CrUX Core Web Vitals on mobile and desktop. Every period is compared with the previous one, with up to 5 years of history.

• Opportunity center

  Drops, low CTR at a good position, lost rankings, content decay, cannibalization and CWV regressions, ranked by criticality. Each item can be dismissed or reopened.

• 49-rule technical crawler

  Crawls up to 25k pages, compares runs and scores AI readiness without mixing up AI search bots and training bots. Nine live tools and a structured-data validator handle one-off checks.

• MCP connector for AI

  More than 80 tools for Claude, ChatGPT and Cursor, through OAuth in Claude or a revocable token with the miranteseo npm package. A Cursor plugin adds a rule for reading the data and two ready-made skills.

• Alerts and shareable reports

  Anomaly alerts, a weekly email summary and report links that open without login and can be revoked at any time. Data can also be exported to CSV.

• 7 free tools

  Site analysis, structured data, robots.txt, page analysis, AI visibility, backlinks and organic traffic. Open to anyone, with no account and no card.

1. 01

   Expensive calls wait for the user's yes

   Connector tools sit in three tiers: reads from already-synced history, live technical checks that cost no money, and paid calls to the SERP, keyword and backlink provider. Paid tools state their price up front, and the most expensive ones show a budget and stop until the person agrees. The conversation gains a step, but an AI assistant cannot run an expensive query on its own.

2. 02

   A repeated question is not paid for twice

   Almost everything bought from the data provider is cached: SERP results for 24 hours; keywords, domains and backlinks for 7 days. The trade-off is that an answer can be up to 7 days old. In return, asking again within that window answers instantly and is not charged again.

3. 03

   The assistant reads the history, not Google

   Read tools answer from data that is already synced, so they don't touch the Google account's daily quota and never need anyone's password. The cost is that the freshest data is whatever the last sync brought in. The few tools that do call Google live, such as URL inspection, are labeled that way.

4. 04

   Database credentials never leave the server

   The miranteseo npm package is just an HTTP client that forwards calls to the same MCP endpoint, authenticated with a revocable mrt_ token. In Claude, authorization goes through OAuth, and revoking cuts access immediately instead of waiting for the token to expire. That means two auth paths to maintain, with the same tools behind both.

5. 05

   Rate limiting doesn't require storing IPs

   The free site analyzer needs no signup and doesn't record who checked what: the rate limit only counts how many analyses came from a connection that day, without storing the IP address. Each block renders as soon as its data arrives and is cached for 7 days (1 day for the technical part), so the same link can be shared. The tool gives up knowing who uses it so it can stay open to anyone.

MCP connector tools

80+

Count from the site's public pages; each tool runs a single query.

technical crawler rules

49

Titles, canonicals, redirects, broken links, robots, HTTPS and AI readiness.

pages per crawl run

25k

Limit on the launch plan.

live technical tools

9

Plus a structured-data and rich-results validator.

free public tools

7

No signup, no card and no record of who checked what.

of stored history

5 years

Maximum history and retention on the launch plan.

help center articles

27

Plus 3 practical guides, in Portuguese and English.

Public numbers from the live site, as of October 28, 2026.