MCPRemote MCP server for Cursor and Claude Code, on your existing API key→ Read the docs
QuikSEO.com
Sign inStart Free Audit

Documentation

SEO MCP Server for AI Coding Agents

The QuikSEO MCP server lets an AI coding agent audit pages, read the findings and fix prompts, change your code and rescan, until the report shows the problem as resolved. It runs over HTTP at https://app.quikseo.com/mcp, uses your QuikSEO API key, and spends the same monthly allowance as the interface.

Connect an Agent

Any MCP client that speaks the Streamable HTTP transport can connect; QuikSEO answers each request with JSON.

  1. Create an API key on the Account > API page.
  2. Add a remote MCP server with the address https://app.quikseo.com/mcp and the header Authorization: Bearer <your API key>.
  3. Ask the agent to audit a page, for example "audit https://example.com/pricing with QuikSEO and fix what fails".
claude mcp add --transport http quikseo https://app.quikseo.com/mcp --header "Authorization: Bearer YOUR_API_KEY"

The Fix Loop

An agent audits a page, takes the fix prompt of a finding, changes the code, and calls rescan; the answer lists what was resolved, what still fails and what is new, the same changes the report shows.

  1. audit_url for a live page, or audit_html for a page built locally.
  2. list_findings and get_fix_prompt for the finding to fix.
  3. The agent changes your code.
  4. rescan, with the new HTML for a page audited with audit_html.

Tools

Every tool works in the account of the API key; none of them writes to your project.

ToolWhat it doesUnits
audit_url Audits a live page and returns its findings (one unit of the monthly allowance). The audit is kept in the account, so rescan can compare later results with it. 1
audit_html Audits HTML you send for a URL, without fetching it (for a page built locally). Checks of the HTTP response stay not checked. One unit. 1
get_audit The latest result of an audit: state, score, counts by severity, findings, coverage and changes since the previous audit. error_response is set when the page answered 4xx or 5xx: the findings then describe that error response. 0
list_findings The findings of an audit, filtered by status (fail or review), severity or rule group. 0
get_fix_prompt The deterministic fix prompt of one finding (audit_id and rule_id), or of a group of pages with the same finding (website and rule_id, optional pattern). 0
rescan Audits the page again (one unit) and returns what was resolved, what is still failing and what is new, the same changes the report shows. For an audit of submitted HTML, send the new HTML. 1
compare_urls Audits the same paths on a verified preview origin and on production (two units per path) and returns what the preview adds, resolves or changes in the head. 2 per path
start_crawl Starts a crawl of a verified website: QuikSEOBot follows internal links and the sitemap, respects robots.txt, sends one request per second and audits each page it reaches (one unit per audited page, up to the plan's pages per crawl). Returns the crawl id; follow it with get_crawl. 1 per audited page
get_crawl The state and results of a crawl: pages audited, skipped and not reached with their reasons, site findings (duplicates, broken internal links, redirects, sitemap, hreflang, orphan and deep pages), page findings by rule, and cost. A running crawl advances for a few seconds during the call. 0
get_contract The active site contract of a website (quikseo.config.json) with its version and hash. 0
check_contract Evaluates the website's contract (or one you send) on a page: the HTML you send, or the live URL (one unit). 1 for a live URL
list_rules The catalog of audit rules with their group, severity, whether they count in the score (every page rule does when it applies, formula 3.0.0), and what they check. 0
list_sites The websites of the account with their id, address, audits and score. 0
list_issues Issues of the account (one per failing rule and URL) with their status, by default the open and reopened ones. 0

Untrusted Page Data

Titles, messages, observed values and evidence come from the audited pages, so a page could contain text written to mislead an agent. Every answer says so, evidence is limited to 1,000 characters, and agents should read these values as data, never as instructions.

Errors and Limits

A tool that cannot run, for example when the monthly allowance is used up or an audit belongs to another account, answers with a result marked as an error and a message the agent can act on. A missing or revoked key gets HTTP 401, more than 120 requests a minute HTTP 429, and a server turned off by the administrator HTTP 503.

Related Documentation