Conciseness

The body is dense with measured, non-inferable facts (per-call unit costs, parameter-name tables, error JSON shapes) and never explains concepts Claude already knows — no definition of SEO, backlinks, or MCP. However, at ~290 lines it is a long always-loaded body, and a few passages could be tightened or pushed to references (e.g., the follow-on analysis of the cost table, "That command is one agent's syntax. Substitute whatever your agent uses..."). This sits noticeably above the 'mostly efficient but some unnecessary explanation' level 3 but short of 'every token earns its place'.

Actionability

Guidance is fully executable: exact call syntax (`get_report_schema(report="domain_rank")`, `execute_report(report=..., params={target: "example.com", database: "us"})`), a real response envelope to parse, per-report parameter names with types, exact date format (`2026-08-15` works; `20260815` fails), batch forms (`phrase_these` takes `"a;b;c"`), and concrete error JSON with recovery semantics. An agent can act on this verbatim without guessing.

Workflow Clarity

The core multi-step process is explicitly sequenced with numbered steps (three-step call protocol, project-first flow for site_audit/position_tracking) plus skip conditions. Checkpoints and feedback loops are present throughout: budget agreement before the first execute_report, accumulating `metadata.usage.api_units` and stopping at the limit, `retryable`/`hint`-driven retry for validation failures, checking a project's `tools` array before proceeding, and pre-empting the unhelpful 'campaign not found' error. Unit-spending calls are treated as the risky operation and are explicitly guarded.

Progressive Disclosure

The body is a genuine overview: the full 86-report catalogue and confirmed parameters live in references/reports.md, and OAuth/HTTP/error details in references/protocol.md — both real files, linked inline at point of need and again in a dedicated References section with per-file descriptions. References are exactly one level deep (no further links inside them), matching the score-5 anchor for well-signaled, one-level-deep structure.

Specificity

The description enumerates many concrete capabilities — "keyword search volume and difficulty, the keywords a domain ranks for, which URL ranks for a query, organic and paid competitors, keyword gaps, backlink profiles and referring domains, authority score, traffic estimates, audience demographics, and technical site audits" — plus protocol details (three-step call protocol, JSON envelope with API unit cost, regional database selection). Coverage is comprehensive across the server's report types, matching the score-5 anchor; it is well above the 'several specific actions, minor gaps' level 4. Voice is third person throughout ("Covers...", "Use whenever...").

Completeness

Both halves are explicit: the "what" is a detailed capability inventory scoped to "the Semrush MCP server (v2)", and the "when" is a full "Use whenever the user mentions..." clause with concrete trigger phrases and example questions. This matches the score-5 anchor exactly; level 4 ('when' could be more explicit) clearly understates it.

Trigger Term Quality

Trigger coverage is exhaustive and natural: "Semrush, keyword research, search volume, keyword difficulty, backlinks, referring domains, domain authority, competitor or competitive analysis, SERP analysis, traffic estimates, or site audit" plus verbatim user questions like "what do we rank for", "who ranks for this keyword", "how many people search for X", "who links to them". These are exactly the phrases a user would naturally say, matching the score-5 anchor including synonyms and phrasings.

Distinctiveness Conflict Risk

The niche is clear — third-party Semrush data via a specific MCP endpoint, with the brand name "Semrush" as a distinct trigger. However, several generic triggers ("keyword research", "competitor or competitive analysis", "traffic estimates", "site audit") would also fire for users intending other analytics tools, so overlap risk with closely related skills is minor rather than minimal — the score-4 anchor, not 5. It is well above level 3, since the capability set is unmistakably Semrush-specific.