Connect your AI.
Bring GeoRanker data into your AI assistant.

GeoRanker SEO & SERP MCP provides search_serps for a query, region and search engine. For example, query car insurance with region Bucharest,Romania, searchEngine google and pages 2 requests up to 20 organic results. Requested depth is capped at 100 organic results; returned coverage can be lower. Results include available ranking positions and source URLs. Multi-keyword domain checks use create_rank_tracking_report instead.

Web Scraping MCP retrieves one supported public page per fetch_page request, including supported JavaScript-rendered and Cloudflare-protected sources. It returns readable text or HTML, capped at 50,000 characters with truncation disclosed. Source access and rendering are not guaranteed. The assistant interprets the content; the MCP does not provide structured field extraction, authenticated browsing or batch site crawling.

The documented integration is a local stdio MCP client launched by your assistant. That client connects over HTTPS to GeoRanker's hosted service. Use a host that can launch local Node.js processes and follow its configuration guide. A cloud-only connector needs a separately supported remote authentication flow; entering a hosted endpoint URL alone is not the documented installation path.

The client enrolls automatically and stores an installation credential and signing identity locally. You do not enter a reusable provider API key or create a GeoRanker account for the Beta trial. The two clients share installation state for the same service origin. This enrollment is not OAuth. The setup check verifies enrollment and tool contracts without submitting a data query; subsequent tool calls use the configured allowance.

Pass forceLive: true to bypass eligible completed MCP cache reuse, which defaults to seven days and is configurable. Matching pending or uncertain work keeps its existing ID. This does not bypass the upstream provider's processing rules or create a background refresh schedule. Preserve cached, cachedAt and generatedAt when supplied; cache storage time is not observation time.

Independent tasks may run in parallel within shared and per-installation service limits. For pending SERPs, retain jobId and call get_serp_result; for pending web pages, call get_fetch_result with the same jobId. SEO report families use their own reportId and matching get tool. Retrieval does not create a new data job. Follow returned nextAction and retry information, preserve pending, unknown and failed states, and avoid duplicate creation. Cancelling or timing out a call does not prove that submitted work stopped.

Report admission units depend on the requested work: keywords × locations × engines for rank checks, selected devices for on-page reports, ceil(maxPages / 100) for broken-link checks (one unit for up to 100 requested pages), one unit for a backlink report, and keyword count for volumes. These are admission weights, not provider credit prices. A schema-valid request can exceed the configured allowance. A create call can count even when it reuses completed data; retrieval has separate limits. Setup and tool discovery do not submit reports. Respect returned limits and retry instructions. For example, three keywords in two locations on one engine require six rank-report admission units. A broken-link request for 20 or 100 pages requires one unit; 1,000 requested pages require ten. None of these values establishes a monetary price or a fixed free quota.

No. The current SEO report contract supports site scope or a single-page check. Path scope is rejected before submission. Site crawls accept depth 1–10 and up to 1,000 requested pages, subject to allowance and provider capacity; defaults are depth 2 and 20 pages. External-link checking is disabled. Report the actual returned internal checks and completion state, rather than claiming complete site coverage.

Retain the report ID and inspect returned validation diagnostics or nextAction. Correct a confirmed rejected input instead of recreating unresolved work. REMOTE_SCHEMA_MISMATCH requires compatible client and hosted contracts before new work. After SCHEDULE_UNCERTAIN, retrieve the same rank report to reconcile the schedule; do not repeat or reverse the change blindly. Unknown completion and HTTP success alone do not establish a completed report.

The client sends requested search arguments or the public URL, an installation identifier, a revocable token and a persisted device signal to GeoRanker over HTTPS. GeoRanker uses its upstream infrastructure to process the request. Local identity and credentials normally live under ~/.config/georanker-search-mcp/client/; GEORANKER_STATE_DIR can change that location. Keep this state private. Shared result caching follows service isolation rules, so avoid secret URLs or sensitive content. Removing the host configuration does not revoke the installation or erase its saved credentials.

The public clients reviewed on 2026-09-18 are version 0.13.1; the website presents both products as Beta. The launcher checks for verified GitHub release packages at startup and every five minutes. It verifies release provenance and installs locked production dependencies without lifecycle scripts. Compatible worker updates apply during an idle period; calls are not replayed. Set GEORANKER_MCP_AUTO_UPDATE=0 in the host environment to review and manage releases yourself. Update downloads contact GitHub, npm and signature-verification services; the client privacy guide explains this data flow.

The reviewed release uses a GitHub source installation: clone, install dependencies, build, run the setup check and configure your host. Its npm packages are unpublished, so an npx command is not available for this release. Requirements are Node.js 22.22.2+ on the 22 release line, 24.15.0+ on the 24 release line, or 26.0.0+; npm and Git. Public source availability does not imply an open-source license; the packages are marked UNLICENSED. Repository metadata is not an official MCP Registry submission.

GeoRanker MCP is in Beta. Request more free Beta credits through the website chat powered by Tidio, or use the contact page. Tell the team which MCP you use, what you are building and the workload you need. Our team reviews your request and confirms the allowance. The hosted service controls shared and per-installation allowances and concurrency, so follow returned limit and retry information. API calculator estimates do not define MCP Beta allowances or report admission units.