GetSEOAgent — Bulk SEO Rewrites for WordPress

Project Info

Client Self-funded SaaS
Year 2026 → present
Category WordPress plugin / SEO
Status Live — v1.7.9 in the WordPress.org directory, free to install
Stack WordPress plugin (PHP 8.1+), admin UI in React 18 + Vite + TypeScript, backend on Bun + Hono (TypeScript) with Postgres, self-hosted via Docker Compose or systemd
Integrations Anthropic Claude, Google PageSpeed Insights, Google Search Console, Rank Math / Yoast / All in One SEO / SEOPress, WooCommerce, Screaming Frog (CSV import), WordPress.org (SVN)
AI Anthropic Claude — an agent loop with 40+ tools; approve a sample of 5 rewrites, then extrapolate across N posts

Project Description

A WordPress plugin that does bulk SEO work through chat, built end-to-end: meta rewrites, schema, redirects, alt text, speed fixes — every one of them a job with history and one-click rollback. It sits on top of whichever SEO plugin the site already runs and writes into its native fields, so sitemaps, schema and on-page analysis keep working. Free on the WordPress.org directory, admin UI in 6 languages, self-hosted backend on your own API keys.

Modules

Module	What it does
Bulk SEO rewrites through chat	Approve a sample of five AI rewrites, then extrapolate the same style across N posts in one job Titles, descriptions and headings rewritten in bulk with a preview before anything is written Style presets and saved prompts — a tone of voice per project, reusable across runs Post, category and tag selection with filters, so a run targets exactly the right slice
SEO-plugin compatibility	Full read + write parity with Rank Math, Yoast, All in One SEO and SEOPress Auto-detects the active plugin and warns when two are running at once Writes into native fields, so sitemaps, schema and on-page analysis keep working Template-type detection so a rewrite matches the kind of page it lands on
Content creation and editing	Creates post and page drafts straight from the chat Proposes and applies content blocks inside existing pages, always preview-first AI FAQ sections generated from the page's own content Site brief and per-post brief keep the project context — the agent can even draft the site brief itself
Speed audit (PageSpeed Insights)	Real PSI runs with Core Web Vitals — LCP, CLS, INP Config-plane fixes plus content fixes: image width/height, lazy-load, WebP через `<picture>` Detects speed optimizers already installed so fixes do not collide Propose → confirm → apply, every change rollback-able
Schema.org	Generates and repairs JSON-LD: Article, FAQPage, HowTo Site-wide schema coverage audit Missing fields backfilled from the site's own data instead of invented Separate branch for WooCommerce product schema
Site plumbing	Repairs broken internal links Collapses redirect chains Fixes broken menu items Hreflang: missing return tags and x-default
Media library alt text	Claude vision writes alt text from the image itself, not from the filename Audit of the media library and galleries: what is missing or junk Bulk apply with rollback
WooCommerce	Product SEO: descriptions and category SEO fields Product schema audit and repair Category-level audits so a whole catalogue branch can be fixed in one run
Screaming Frog import	Upload a crawl CSV from Screaming Frog The agent fixes what it honestly can and flags the rest Findings stored as an audit that can be queried from chat
Google Search Console	Connects GSC and queries the data from inside the chat Attribution: which change moved which page and query Property matching and URL canonicalization so the data lines up with the site
Jobs, history and rollback	Every bulk run is a job with status and progress Mid-flight cancel A per-job audit log of what exactly was written One-click rollback of a whole run
Keys, privacy and self-hosting	Self-hosted backend — Docker Compose or systemd, on your own infrastructure BYO keys for Anthropic and Google PageSpeed Insights, encrypted at rest in a vault Per-site license registry and tier gating The site's data never leaves the pair of endpoints the user controls
Cost metering and guardrails	Every LLM call metered: tokens in / out / cache and cost in micro-USD Credit pool with caps and throttling before a run can burn a budget Anomaly detection with alert emails when spend goes off the rails Per-model pricing table so cost is attributed correctly as models change
Distribution and tiers	v1.7.9 in the WordPress.org directory, installs free Free tier with BYO key, plus Pro, Agency and Enterprise Recurring billing through WayForPay with top-up packs Affiliate cabinet on the marketing site
Admin UI	React 18 + Vite app embedded into the plugin 6 languages: EN, RU, UK, ES, FR, PT-BR Chat, job list, history and audits in one panel — no leaving wp-admin Streaming answers over SSE so a long run is visible as it happens

Architecture

Four components in one repo, three toolchains: the backend (Bun + Hono, TypeScript) holds the agent loop, the 40+ tools, licensing, credits and metering; the plugin (PHP 8.1+) owns everything that writes to a user's site — REST controller, SEO-plugin adapters, job store, history, rollback, redirects; the admin app (React 18 + Vite) is embedded into the plugin; the landing pages are kept in the same tree. Product logic lives in the backend, site mutations only ever go through the plugin. Every LLM call is metered into a usage table (tokens in/out/cache plus cost in micro-USD), which is what makes cost per job and anomaly alerts possible. The plugin ships to wp.org over SVN; the backend runs from a systemd unit with no build step. Tests: PHPUnit on the PHP side, bun test on the backend, vitest in the admin app — plus a loader-parity test that fails the build when a new class is not required in the bootstrap file.

Visit Live SaaS