Published · Updated · Reviewed by the OnRankSEO team
An MCP server, a Claude plugin and a ChatGPT app. OnRankSEO’s tools run a real audit: it crawls the site against 121 audit checks, records each finding with what was seen, where and when, and the published rule it rests on, and drafts a report that a person at OnRankSEO checks and signs.
These limits are the reason to use it. An assistant connected to OnRankSEO can only pass on what OnRankSEO can stand behind.
The server’s address is https://onrankseo.com/api/mcp. There is no API key: you sign in with the OnRankSEO account you already have.
claude mcp add --transport http onrankseo https://onrankseo.com/api/mcp; in ChatGPT, as an app or connector by its MCP address.You can end a connection at any time, in either of two places: remove OnRankSEO from your assistant’s connector settings, or end every connection to your account at onrankseo.com/auth/connections — that signs your account out everywhere, and the assistant’s next call is refused.
/plugin marketplace add onrankseo/seo
/plugin install onrank@onrankseoSkills: /onrank:foundations, /onrank:first-look, /onrank:audit, /onrank:report, /onrank:link-plan, /onrank:local-grid, /onrank:ai-visibility, /onrank:evidence. foundations loads by itself. A subagent, onrank-auditor, runs long audits.
In Claude on the web and desktop, ask in plain words: “give me a first look at example.com”, “draft the report for my client’s audit”.
| Tool | What it does | Who |
|---|---|---|
| first_look | Twelve page checks on one page and what the site tells eight AI systems' crawlers — each line with what we found, what it means, what to do and the published rule. | Any account |
| checks_catalogue | Every check the OnRankSEO crawler runs on each page: plain name, severity, what it means, how to fix it, who does it, difficulty and SEO impact, and the rule it rests on: a page in our source register, named with its publisher and the date we read it, or 'our standard'. | Any account |
| score_formula | The published OnRank Score: every input, its weight, the rule behind it and where it is read from. | Any account |
| run_audit | Opens (or reuses) the client's in-progress audit and gathers the market data: the full crawl and check set, and the licensed data where keys are set. | Console account |
| audit_status | An audit's status, due date, what has been gathered and when, how many findings are recorded, the pre-ship gate, and whether it has been delivered (by a person). | Console account |
| get_findings | Every finding recorded on the audit, in the order the report prints them, each with what we saw, where and when, and the rule it rests on (named, quoted only where the quotation is held verbatim, or flagged as our observation). | Console account |
| add_finding | Records a finding on the audit, validated exactly as the console validates it. | Console account |
| evidence | The full evidence for one finding (by its number or letter as the report prints it): what we saw, where, the date, the rule, our dated measurements and any third-party reading, named as theirs. | Console account |
| pending_access | Every check that could not run yet, the access that unlocks it (never a password), the steps, and what we will report once it runs. | Console account |
| observe_competitors | Reads the client's site and up to three competitors with our crawler, one after another, and stores every cell with today's date for the report's comparison table. | Console account |
| gather_market | The full crawl with the 121 audit checks, keywords, gaps, traffic and backlink estimates (licensed data, labelled estimate with source and month), and speed — each section stored, or 'not measured — needs <KEY>'. | Console account |
| map_grid | A 5 × 5 grid around the client's address: its position in Google Maps results for one search at each point (Places API), dated, with the month-over-month change. | Console account |
| link_check | From the stored crawl: distinct broken internal targets and the pages linking to them, dead outbound targets (one fix when it is a shared template link), and the robots-blocked addresses by kind and rule. | Console account |
| build_report | Builds the report draft from the record, as the console renders it: the findings with evidence and rule, what is pending, the plan, and the console address where a person checks the 24-point gate and signs it. | Console account |
| report_sections | Each section of the report: stored with its month, or 'not measured — needs <KEY>' with the key or step that unlocks it. | Console account |
| gate_status | The pre-ship gate for the audit's deliverable: how many of the 24 items a person has passed, which failed or are unchecked. | Console account |
| link_plan | Which of the ten published link-building tactics apply to this client and why (from the record), what we would do, who does it, the week and what each cannot promise — with the two standing cautions. | Console account |
| dealer_pack | The structured data the site lacks, a ready JSON-LD block built from the site's own pages (each field naming its page), the Business Profile checklist with Google's help pages, and whether the name, address and phone agree. | Console account |
| score | The score from the published formula, its three sub-scores, and every input not in it yet with what it needs. | Console account |
| lab_readings | The Lab's stored weekly readings for one site: Google and Bing positions per search, and AI mentions as runs and named-in counts — each dated, with its source. | Console account |
| lab_tactics | The Lab's recorded experiments matching a search: what was changed, the measure, before and after with their source, and the verdict a person recorded (worked, no effect, hurt, stopped). | Console account |
| queue_email | Queues a draft email in the Approvals queue. | Console account |
Sent to the assistant on every connection, and written into every skill: