Audit the site you shipped. Name the code that produced it.
Weavatrix SEO is a Search Evidence Graph, not a detached crawler report. Live HTTP, repository routes, schema, claims, and observations bind to one revision. CLI and MCP are the same binary. Missing evidence stays unmeasured.
Install from npm Add the MCP host ↓# one native binary: CLI commands and MCP tools $ npx -y weavatrix-seo audit --site https://example.com --json $ claude mcp add weavatrix-seo -- npx -y weavatrix-seo mcp --allow-root . $ cargo install weavatrix-seo-cli --locked
Why this is not another crawler
A crawler scores pages. Weavatrix SEO names the producer. A finding on /iphone-repair/haifa can point at the Next.js helper that emitted the title. Silence never turns green.
Code-aware
Next.js App/Pages, Nuxt, and Astro families are predicted from source. Hybrid mode classifies SOURCE_ONLY versus RESPONSE_ONLY against the crawl budget.
Evidence-honest
Authority, method, and scope sit on every finding. A Google rich-result miss is not a schema.org vocabulary miss. A target outside the crawl is UNMEASURED, not healthy.
Agent-native
Fifteen MCP tools match the CLI. An agent can query orphans, retrieve a service page, and explain a fingerprint without a shell.
Read-only
SEO never writes pages, never generates articles, never applies patches. seo_plan can hand a location to Weavatrix Refactor; SEO still does not mutate source.
CLI commands
The binary is weavatrix-seo. weavatrix-seo mcp is the agent socket, not a second product. Every command takes --json.
audit
Crawl + findings by axis, severity, and authority. --ci / --baseline / --html.
inventory
Measured URLs, routes, producers. No ranking.
opportunities
Gaps to build, not current errors.
plan
Target search architecture and a read-only Refactor handoff.
compare
Structural gaps vs public competitors. Prose is never copied.
query
Bounded FROM urls WHERE inbound_links = 0 LIMIT 20.
retrieve
Lexical candidate pages. semantic stays null for the shipped model.
explain / diff
URL → route → symbol chain. Snapshot and worktree deltas.
15 MCP tools — same engine
Point Claude, Codex, or Cursor at npx -y weavatrix-seo mcp --allow-root . MCP default is --public-only.
seo_audit
Findings with authority. Google rich-result ≠ schema.org vocabulary.
seo_inventory
The measured search surface: URLs, routes, producers.
seo_opportunities
Construction work, not current errors.
seo_plan
Target search architecture. Read-only Refactor handoff.
seo_gate
CI verdict instead of an exit code.
seo_compare
Locale, schema, FAQ, cardinality, H1 gaps.
seo_query
Orphans, noindex, missing H1 — without dumping the crawl.
seo_retrieve
Lexical candidate pages. Inferred, not exact.
seo_chunks
Answering chunks from first-party pages.
seo_similar
Near-duplicate first-party pages.
seo_explain
One fingerprint to the source helper when the repo is in scope.
seo_diff
Snapshot, audit JSON, or worktree deltas.
seo_links
Directed internal links from first-party page vectors.
seo_vectors
First-party lexical vectors. No embedding API on the default path.
seo_observations
GSC, logs, and AI-citation observations when you import them.
# Codex ~/.codex/config.toml [mcp_servers.weavatrix-seo] command = "npx" args = ["-y", "weavatrix-seo", "mcp", "--allow-root", "."] # Cursor mcp.json # "weavatrix-seo": { "command": "npx", "args": ["-y", "weavatrix-seo", "mcp", "--allow-root", "."] }
Read-only. SEO does not write pages or apply patches. Off-crawl canonicals are UNMEASURED. Mixed content is subresources, not navigation. GitHub · npm · crates.io