Build with SiteTR.
One MCP server, 18 grounded tools, one flat price. Get from install to your first audit in about four minutes.
Menu
Quickstart
SiteTR ships as a SaaS dashboard and as an MCP server. This page gets you from zero to a prioritized fix plan in your AI agent.
1. Install the MCP server
SiteTR exposes 18 grounded tools over the Model Context Protocol. Add the server to your agent config once, and every tool is available to Claude, Cursor, or any other MCP-compatible client.
Where the token comes from
The bearer token is an installation-level secret — one token per SiteTR instance, issued by whoever operates it. There is no signup form for it.
The token grants read access to the instance’s data. Write tools stay off unless the operator turns them on — submit_audit requires MCP_ALLOW_SUBMIT=true on the server.
2. Connect Search Console (once, in setup)
SiteTR’s value is that findings are cross-referenced with your real search performance — not a black-box difficulty score. Connect the Search Console property whose traffic you actually care about once, in the SiteTR dashboard — every tool that needs it (Decay Scan, audits, Content Gap) reads it from then on.
3. Ask a real question
No prompt-crafting required. SiteTR is built for the questions you’d ask a senior consultant on their first day.
The tools
Every tool follows the same evidence/verify contract. Tool descriptions are what sells an MCP server — the ones in your client are these, verbatim.
capabilitiessite_overviewlatest_auditdomainrequired — Site domain, e.g. example.comrank_historydomainrequiredkeyword_researchseedrequiredkeyword — Alias for seedglhllimitalerts_listdomain — Optional: restrict to one sitelimitsubmit_audittargetrequired — Absolute URL or domain to auditdomain — Alias for targetmax_crawl_pages — 1-25, default 10mode — Default fullaudit_statusdomainrequiredaudit_issuesdomainrequiredseverityaudit_sitedomainrequired — Site domain, e.g. example.comgeo_readinessdomainrequiredlive — Re-probe now (robots + crawl sample + bot probe) instead of reading the latest audit. Default false.decay_scandomainrequiredmonitor_driftdomainrequiredlimitkeyword_clusterskeywordsrequired — 1-50 keywords to clusterglhldomain — Optional: flag clusters where this domain already competescontent_gaptargetrequired — Your site (URL or domain)competitor — Optional: skip discovery and diff against this siteseed_keyword — Required when competitor is not given — seeds SERP discoveryglhlgenerate_metaurlrequired — The page URL (brand and canonical derive from it)keyword — Target keyword for the primary variantspage_title — Existing title, kept as one variantpage_description — Existing description, used as the base when giventone — Optional LLM tone refinement (e.g. 'professional', 'playful') — costs a rate-limited LLM callgenerate_schematyperequired — Schema type, e.g. Organization, LocalBusiness, FAQPageurlrequiredpage_titleentity_reportdomainrequiredname — Brand name to resolve (defaults to the site's derived name)force — Bypass the cache and re-check live sourcesLimits, transport & errors
Per-tool ceilings, generated from the server’s own rate-limit table — the expensive tools throttle, cheap reads don’t. Limits are per instance and reset on a sliding window.
Tools not listed have no per-tool ceiling — they read the instance’s stored data and stay cheap.
Evidence & verify
Everyone worries AI agents make things up. SiteTR’s evidence/verify pattern is exactly what agent tooling needs in 2026: the agent advises, the engine proves.
Every finding ships with three things
Evidence — the specific selector, header, request, or measurement that produced the finding. Verify — a copy-paste step (curl, DOM query, or agent tool call) that confirms the fix worked. Provenance — which provider the data came from and when it was last fetched. Never a fake number.