Bulk AI CLI Connector: Edit Pages, SEO and Schema with Claude, ChatGPT and MCP Agents

By puneetindersingh

60+ active installationsVersion 2.6.2

Connect Claude, ChatGPT and MCP agents to WordPress. Bulk edit SEO meta, schema, redirects and Elementor page copy over the REST API. With undo.

Description

Bulk AI CLI Connector connects AI agents and command-line tools to a WordPress site. Claude, Claude Code, ChatGPT, an MCP client or a plain script can read and rewrite SEO titles, meta descriptions, JSON-LD schema, redirects, image alt text and the visible copy of a page, in bulk, over an authenticated REST API. Every page content edit is backed up first and can be undone.

There is no AI inside the plugin and it sends nothing anywhere: it makes no requests of its own to any service and needs no AI account. It gives the agent you already use a safe way into the site. Each call signs in as a WordPress user and can only do what that user could do in wp-admin.

See what is live, where you work

New in 2.3: an SEO item in the admin bar on every page you can edit. It opens a panel over the page with what the page really sends to search engines (title, description, robots, canonical, social tags, headings, every JSON-LD block), read from the page itself, so it is true whichever plugin printed it. Problems are marked, the title and description can be changed in place, custom schema can be added, and a history lists every SEO change on the page with who made it and how: in the editor, through an API key, or from the panel. The block editor and the classic editor get the same live view beside the post.

What you can do with it

  • Connect Claude or ChatGPT to WordPress: The plugin is its own MCP server. claude.ai, the Claude desktop app and ChatGPT add it as a custom connector with one address and an Allow click on your site. Claude Code, Cursor, VS Code and other MCP clients add the same address with a key. Nothing to install on the computer. Agents that do not speak MCP call the REST routes directly. Anything else that can send an authenticated HTTP request works the same way, including n8n, Zapier and Make.
  • Bulk edit Yoast SEO and Rank Math meta: SEO titles, meta descriptions, focus keywords, canonical URLs, robots directives, Open Graph and Twitter fields on up to 100 posts, pages, products or term archives in one call. Works on the free editions.
  • Edit Elementor and Beaver Builder pages with an AI agent: read the headings, paragraphs, button labels and list items of a page as a list, change only the ones you name, and leave the layout exactly as it was. SiteOrigin Page Builder, Advanced Custom Fields text fields, Gutenberg, the classic editor, Divi and WPBakery pages are covered too.
  • Add JSON-LD schema: FAQPage, HowTo, Service or any other type, per post or per category, merged into the Yoast or Rank Math schema graph.
  • Add redirects: 301, 302, 307 and 410 rules, saved into Rank Math’s Redirections where that is in use.
  • Bulk update image alt text: up to 200 media library images per call, matched by image URL.
  • Work in a spreadsheet: export every SEO title and description as CSV, edit in Excel or Google Sheets, import it back.
  • Undo: every page content write stores the previous version first. Dry runs show what would change and write nothing.

Works with

  • SEO plugins: Yoast SEO and Rank Math, free and paid editions. The active one is detected automatically
  • Page builders and editors: Elementor (including Elementor 4 atomic widgets), Beaver Builder, SiteOrigin Page Builder, Advanced Custom Fields, Gutenberg, the classic editor, Divi, WPBakery
  • WooCommerce: products, product categories and product tags, plus return policy and shipping details on Product schema
  • Agents and tools: Claude Code, Claude Desktop, ChatGPT, MCP clients, n8n, Zapier, Make, cURL, Python and Node scripts, headless front ends

Why it exists

WordPress does not expose Yoast SEO or Rank Math fields over the REST API, and page builders keep a page in post meta where the core REST API cannot reach it. Editing a meta title or an Elementor heading programmatically normally means WP-CLI with server access, a custom plugin, or a premium tier. This plugin adds a clean, authenticated REST surface instead.

It auto-detects which SEO plugin is active and exposes plugin-neutral field aliases (title, description, focus_kw, …) so you don’t need to memorise Yoast vs Rank Math meta keys.

This plugin was called Bulk SEO Meta Editor for AI Agents up to version 1.9.1. Same plugin, same address, same REST routes, new name: it now edits page content as well as SEO fields.

New in 2.0: page copy, redirects and API keys

Version 2.0 goes past SEO fields. The same REST API can now read and change the visible text of a page, wherever the site keeps it:

  • Page builders: Elementor (including Elementor 4 atomic widgets), Beaver Builder, SiteOrigin Page Builder, and text fields in Advanced Custom Fields. These keep a page in post meta, where the core REST API cannot reach it. The plugin lists every piece of text as a node and changes only the ones you name. Layout, styling and widgets it does not know are written back exactly as they were read. In Advanced Custom Fields it can also add a row to a repeater or flexible content field (an FAQ item, a product tab), and edit the ACF fields of a category or any other taxonomy term.
  • Gutenberg, classic editor, Divi and WPBakery: read and replace the post content, or run a literal find and replace across it. From v2.6.2 this covers Divi Theme Builder headers, footers and body layouts too, for accounts that may change the theme, edit others’ posts, post unfiltered HTML and use the Theme Builder.
  • Term descriptions: category and product category introductions keep their headings, lists and tables.
  • Undo for every edit: each write stores the previous version first. GET /content/backups lists what can be restored and POST /content/revert puts it back. The newest ten are kept per page.
  • Dry run and conflict check: dry_run shows what would change and writes nothing. base_hash refuses a write when the page changed since it was read.
  • Redirects: POST /redirect adds a 301, 302, 307 or 410 rule. On a site that uses Rank Math’s Redirections the rule goes into Rank Math itself. Elsewhere the plugin serves it.
  • Copy a page: POST /clone copies a page with its template, design meta and featured image into a new draft.
  • Merchant listing fields: fills in return policy, shipping details and brand on WooCommerce Product schema from one stored policy.
  • API keys: for sites where a security plugin or the host blocks application passwords. Made on Settings > AI CLI Connector, shown once, stored only as a hash.

What’s included

  • REST endpoints: read and write SEO meta on any post, page or custom post type via the standard /wp/v2/posts/{id} route (reads need ?context=edit) or the namespaced helpers
  • Taxonomy term archives: edit SEO meta on category/tag/custom-taxonomy archive pages too, not just posts (Yoast and Rank Math both supported)
  • Bulk update: apply changes to up to 100 posts or terms in a single call
  • Bulk image alt text: update _wp_attachment_image_alt on up to 200 media-library attachments per call by image URL. Resolves CDN paths and -WIDTHxHEIGHT thumbnail suffixes back to the parent attachment automatically (v1.5.0+)
  • CSV import / export: round-trip your SEO meta through Excel or Google Sheets, posts and terms in one file
  • MCP server built in: one address on your site for Claude Code, Cursor, VS Code and other MCP clients, with 15 tools for SEO fields, schema, alt text, page copy, undo, page copies and redirects
  • Auto-detection: works with Yoast SEO or Rank Math, picks the active one automatically
  • Per-object permission checks: Contributors and Authors can only edit posts, media and terms they could edit in wp-admin; site-wide archive and global settings require an administrator
  • Allowlist enforcement: only meta keys belonging to the active SEO plugin are accepted; arbitrary postmeta writes are rejected
  • Custom JSON-LD schema (v1.6.0+): store structured data (FAQPage, HowTo, Service, or any type) per post, and since 1.9.0 per category, tag or custom taxonomy term; it is injected into Yoast’s or Rank Math’s existing schema graph rather than as a duplicate block. When no SEO plugin schema graph is printed on the page (no Yoast or Rank Math, or their schema output switched off), the plugin prints the stored schema as its own JSON-LD block (v1.9.0+)
  • Post titles (v1.9.0+): send post_title in a /bulk item to rename a post; the slug is never touched
  • Sitemap refresh (v1.9.0+): POST /sitemap/flush clears the Yoast or Rank Math sitemap cache after REST edits
  • Invalid Product schema guard (v1.9.0+): drops Product nodes that carry no offers, review or aggregateRating, the ones Search Console reports as invalid on quote-only WooCommerce catalogues. Disable with add_filter('bulkseme_drop_invalid_product_schema', '__return_false')
  • Page content (v2.0.0+): read and edit the text of Elementor, Beaver Builder, SiteOrigin, ACF, Gutenberg, classic, Divi and WPBakery pages and of term descriptions, with a backup before every write. From v2.1.0: add rows to ACF repeaters and flexible content, and edit ACF fields on terms. From v2.2.0: Elementor dropdowns and switches, and adding a widget by copying an existing one
  • Redirects, page copies and merchant listing fields (v2.0.0+): see “New in 2.0” above
  • API keys (v2.0.0+): a second way to sign in, next to application passwords

REST endpoints

All endpoints live under /wp-json/seo-meta-bridge/v1/ (or ?rest_route=/seo-meta-bridge/v1/... on plain-permalink installs):

  • GET /status: detect the active SEO plugin and list available fields
  • POST /bulk: update SEO meta on up to 100 posts/terms in one call
  • POST /bulk-alts: update image alt text on up to 200 media-library attachments by URL
  • GET /export: export all posts’ SEO meta as CSV, returned in the JSON response body
  • POST /import: apply updates from a CSV upload or JSON rows array
  • GET /schema (v1.6.0+): read back the custom JSON-LD schema stored on a post; add kind=term&taxonomy=<slug> for a term (v1.9.0+)
  • POST /sitemap/flush (v1.9.0+): clear the Yoast or Rank Math sitemap cache so the next sitemap request rebuilds from live data
  • GET /content?id=N (v2.0.0+): which builder owns the page and its text as nodes; add kind=term&taxonomy=<slug> for a term description
  • POST /content (v2.0.0+): change nodes with updates: [{el, setting, value}], or send post_content for a page that keeps its text there
  • POST /content/replace (v2.0.0+): literal find and replace across the page’s text
  • GET /content/backups and POST /content/revert (v2.0.0+): list and restore earlier versions
  • POST /clone (v2.0.0+): copy a page or post into a new draft
  • GET, POST, DELETE /redirect (v2.0.0+): list, add or remove redirect rules
  • GET, POST, DELETE /merchant-policy (v2.0.0+): the return policy and shipping details added to WooCommerce Product schema
  • POST /register-key (v2.0.0+): store the hash of an API key from your own signed-in browser session
  • GET, POST /update (v2.0.2+): see which version is installed and which one WordPress.org offers, and install the offered version without opening wp-admin

The standard WordPress REST route also works: POST /wp/v2/posts/{id} with a meta payload containing Yoast or Rank Math keys.

Field aliases

Alias
Yoast meta key
Rank Math meta key

title
_yoast_wpseo_title
rank_math_title

description
_yoast_wpseo_metadesc
rank_math_description

focus_kw
_yoast_wpseo_focuskw
rank_math_focus_keyword

canonical
_yoast_wpseo_canonical
rank_math_canonical_url

og_title
_yoast_wpseo_opengraph-title
rank_math_facebook_title

og_image
_yoast_wpseo_opengraph-image
rank_math_facebook_image

tw_title
_yoast_wpseo_twitter-title
rank_math_twitter_title

Plus robots, OG/Twitter description fields, and Rank Math primary-taxonomy keys.

MCP server built in

The plugin answers MCP requests itself at /wp-json/seo-meta-bridge/v1/mcp (Streamable HTTP). Nothing to install on your computer.

  • claude.ai, the Claude desktop app, ChatGPT: add a custom connector with that address. The app opens a page on your site; sign in as an administrator and click Allow. This is OAuth 2.1 with PKCE, and the site must be public and on HTTPS.
  • Claude Code, Cursor, VS Code and others: make a key on Settings > AI CLI Connector and copy the ready setup, or a message for any other AI tool.

It has 15 tools: site status, SEO export and bulk update, stored schema, image alt text, read and edit page content, find and replace, content backups and undo, page copies, redirects and sitemap rebuilds. Each tool runs one of the plugin’s own REST routes with the same permission checks, so a key limited to this plugin stays limited. A one-click connection works on the MCP server only, never on other REST routes, and is listed under Connected apps with a Disconnect button.

The older Node.js MCP server (bulk-seo-meta-editor-mcp on npm) still works for Claude Desktop.

Security

  • Application Passwords (Basic Auth), HTTPS required in production. API keys (v2.0.0+) work over the REST API only, over HTTPS only, on this plugin’s routes only unless made with the wider scope, and are stored as SHA-256 hashes
  • Per-post edit_post capability check on every read and write, edit_term for terms, manage_options for redirects and site-wide settings
  • The SEO plugin’s and the page builder’s own rules for who may edit are followed as well: Yoast and Rank Math role settings, Elementor’s excluded roles, Beaver Builder’s Builder Access setting
  • Page copy is always cleaned with the post content allowlist (wp_kses_post), whoever sends it. Scripts and event handlers are never stored
  • Meta keys are allowlisted to the active plugin’s known fields
  • URL-shaped fields (canonical, og_image, twitter_image, link URLs, redirect destinations) sanitised through esc_url_raw
  • One admin screen, Settings > AI CLI Connector, and only administrators can open it

Trademarks

Yoast SEO is a trademark of Yoast BV. Rank Math is a trademark of RankMath. Claude is a trademark of Anthropic PBC. ChatGPT is a trademark of OpenAI OpCo. WordPress is a trademark of the WordPress Foundation. All other product names, logos, and brands are property of their respective owners. This plugin is an independent integration and is not affiliated with, endorsed by, or sponsored by any of them. Trademarks are referenced only to describe interoperability with the listed products.

Installation

  1. Install from the WordPress.org plugin directory: Plugins → Add New, search for Bulk AI CLI Connector. Or with WP-CLI: wp plugin install bulk-seo-meta-editor-for-ai-agents --activate. Or upload bulk-seo-meta-editor-for-ai-agents.zip via Plugins → Add New → Upload Plugin
  2. Activate. A notice offers Set up now, or click Settings next to the plugin on the Plugins screen. Both open Settings > AI CLI Connector
  3. Click Make my key, pick your AI tool and copy its setup: a Claude Code command, an MCP settings block for Cursor and VS Code, or a message for any other AI tool. An Application Password (Users > Your Profile > Application Passwords) works too where the host allows it
  4. Test: curl -u 'username:app pass' https://yoursite.com/wp-json/seo-meta-bridge/v1/status, or with a key: curl -H 'X-Bulkseme-Key: bsme_...' https://yoursite.com/wp-json/seo-meta-bridge/v1/status

MCP clients (Claude Code, Cursor, VS Code): use the setup the settings screen gives you after you make a key. It adds /wp-json/seo-meta-bridge/v1/mcp with the key as an Authorization: Bearer header.

FAQ

How do I connect claude.ai or ChatGPT without a key?

Open Settings > AI CLI Connector, copy the address from the “no key needed” box, and add it as a custom connector in Claude or ChatGPT. Your AI app opens a page on your site: sign in as an administrator and click Allow. The site must be public and on HTTPS. Connected apps are listed on the same screen with a Disconnect button, and one-click sign-in can be switched off there. Please report a security problem privately through “Report a vulnerability” on the plugin’s GitHub page (the Plugin URI), never in a public forum post.

How do I connect Claude or ChatGPT to WordPress?

Install and activate the plugin, then open Settings > AI CLI Connector (or click Settings under the plugin’s name on the Plugins screen), and either copy the MCP address into claude.ai or ChatGPT as a custom connector, or make a key and copy the setup for your agent. An Application Password under Users > Your Profile works too where the host allows it. Claude Code, Cursor and VS Code connect to the plugin’s built-in MCP server with the setup the screen gives you. Other agents call the REST routes under /wp-json/seo-meta-bridge/v1/ directly. ChatGPT and other tools work wherever they can send an authenticated HTTPS request. Start with GET /status, which reports the active SEO plugin and the fields available.

Is this an MCP server for WordPress?

Yes. Since 2.5.0 the plugin is an MCP server itself, at /wp-json/seo-meta-bridge/v1/mcp, with 15 tools covering SEO fields, schema, alt text, page content with undo, page copies and redirects. Everything is also a REST route, which an agent can call over HTTPS without MCP.

Do I need WP-CLI, SSH or server access?

No. “CLI” in the name refers to the command-line agents and scripts on your own computer. Everything goes over HTTPS to the WordPress REST API, so it works on shared hosting and managed hosts where you have no shell.

Does the plugin send my content to an AI service?

No. The plugin contains no AI and makes no requests of its own to any service. It only answers REST calls from a signed-in WordPress user. (The one exception is not about content: when asked through /update, it has WordPress check WordPress.org for a new version of this plugin and, if told to install it, download it from there.) What your agent does with the text it reads is between you and that agent.

How do I bulk edit meta descriptions in WordPress?

Activate the plugin, create an Application Password under Users → Your Profile, then send one POST to /wp-json/seo-meta-bridge/v1/bulk with up to 100 items. Each item is a post ID plus the fields to change, for example {"id":123,"meta":{"description":"..."}}. If you would rather work in a spreadsheet, GET /export returns every row as CSV (with live character counts) and POST /import writes the edited file back.

Can I bulk edit Yoast SEO titles without Yoast Premium?

Yes. The plugin writes the same postmeta keys Yoast Free reads from, so bulk title and description edits work on the free edition. Rank Math Free works the same way.

Does it work with Rank Math as well as Yoast?

Both. It detects whichever plugin is active and maps plugin-neutral aliases (title, description, focus_kw) onto the right meta keys for that plugin, so the same request body works on either site.

Can Claude or ChatGPT edit my WordPress SEO meta?

Yes, that is what it was built for. Through the built-in MCP server, Claude Code and other MCP clients read and bulk update titles, descriptions, focus keywords, canonicals, social fields and schema. Any other agent or automation tool can call the REST endpoints directly with a key or an application password.

How do I bulk update image alt text?

POST /bulk-alts accepts up to 200 { image_url, new_alt } pairs per call and writes _wp_attachment_image_alt on the matching media-library attachment. CDN paths and -WIDTHxHEIGHT thumbnail suffixes resolve back to the parent attachment automatically.

Can I add custom JSON-LD schema to a post?

Yes, since 1.6.0. Schema is stored per post and injected through Yoast’s wpseo_schema_graph filter or Rank Math’s rank_math/json_ld filter, so your nodes merge into the graph those plugins already output rather than appearing as a duplicate second block. Add mode appends, replace mode swaps the graph.

Can I add schema to a category or product category archive?

Yes, since 1.9.0. Send a /bulk item with kind: "term", the taxonomy slug and a schema field, the same way you send term titles and descriptions. The schema renders on the first page of that term’s archive only, which is the page that shows the term description. Read it back with GET /schema?id=<term_id>&kind=term&taxonomy=<slug>. Writing needs the edit_term capability on that term.

Does the schema work without Yoast or Rank Math?

Yes, since 1.9.0. If no SEO plugin schema graph is printed on the page, whether because neither plugin is installed or because its schema output has been switched off, the plugin prints the stored schema as its own <script type="application/ld+json"> block. Meta title and description editing still needs Yoast or Rank Math, because those fields belong to them.

Why did the Product schema disappear from some WooCommerce product pages?

Since 1.9.0 the plugin removes Product nodes that carry none of offers, review or aggregateRating from the Yoast and Rank Math schema graphs. Google requires at least one of them, so a quote-only product (no price) can never earn a product rich result and only produces an “invalid” entry in Search Console. Products with a price are untouched, and the node comes back on its own once a price is added. To keep those nodes anyway: add_filter('bulkseme_drop_invalid_product_schema', '__return_false');.

Can I rename a post through the API?

Yes, since 1.9.0. Send post_title in a /bulk item. The value is sanitised as plain text, the slug and URL never change, and the caller needs edit_post on that post. /import ignores the post_title column, so a CSV round trip cannot rename anything.

Does this work with a headless or decoupled WordPress setup?

Yes. Every field is readable and writable over REST, so a headless front end, a CI job or a content-ops script can manage SEO meta the same way it manages any other structured field.

Does this work with Yoast SEO Free / Premium / Rank Math Free / PRO?

Yes. The plugin uses the postmeta keys that all editions read from, so it works regardless of which tier is installed.

Can I run both Yoast and Rank Math at the same time?

You shouldn’t. They conflict with each other in the front-end (duplicate meta tags). This plugin auto-detects whichever is active; if both are active, Yoast wins.

Is it safe to install on production?

Yes. The plugin does nothing until something calls the REST endpoints, and every call needs a signed-in user who could make the same change in wp-admin. SEO fields, schema, alt text and titles are checked per post. Since 2.0.0 page copy can be edited too: each write is cleaned with the post content allowlist, stores a backup first, and can be reverted. The plugin cannot create users, change settings outside its own, or run code that was sent to it. The only thing it can install is its own update from WordPress.org (through /update, for users who may update plugins).

Can Claude or ChatGPT edit an Elementor or Beaver Builder page?

Yes, since 2.0.0. GET /content?id=<page id> returns the page’s headings, paragraphs, button labels and list items as nodes, each with the element id and setting name it lives in. Send the ones you want changed to POST /content as updates: [{"el": "...", "setting": "...", "value": "..."}]. Only those values change. Add "dry_run": true to see the result without writing.

How do I undo an edit?

Every write through /content stores the previous version first and returns its id as backup_ts. POST /content/revert with the page id and that ts restores it, and the revert can itself be undone. GET /content/backups?id=<page id> lists the newest ten. A backup can only be restored onto the page it was taken from.

Who can edit page content through the API?

Only a user who could edit that page in wp-admin. The plugin checks edit_post on the page (or edit_term on the term) for every read and write. Where the page builder has its own rule it is followed as well: a role that Elementor excludes from its editor, or that Beaver Builder’s Builder Access setting keeps out, is refused here too.

Can I send scripts or iframes in page copy?

No. Every value is cleaned with WordPress’s post content allowlist (wp_kses_post) before it is stored, whoever sends it, administrators included. A site owner who needs the unfiltered behaviour WordPress gives administrators in the editor can switch it on with add_filter('bulkseme_allow_unfiltered_html', '__return_true');. Even then the caller must hold the unfiltered_html capability and the request must ask for it with unsafe_html: 1.

How do API keys work, and are they safe?

An administrator makes a key on Settings > AI CLI Connector. It is shown once and only its SHA-256 hash is stored, so a copy of the database holds nothing that can be replayed. A key signs REST API requests in as the account that made it, with that account’s capabilities, and is removed when that account is deleted. It is read only for REST API requests over HTTPS, so wp-login.php, wp-admin, admin-ajax and XML-RPC never see it.

By default a key is accepted on this plugin’s own routes and nowhere else (and not on /update, which installs a new version of the plugin). If a tool also needs WordPress’s own REST routes (to list posts or upload media, for example), make the key with the wider scope. Such a key is as powerful as an application password for your account, so store it like a password. Revoke any key on the same screen. To switch key sign-in off entirely: add_filter('bulkseme_api_key_auth_enabled', '__return_false');.

How do I keep the plugin up to date on sites I manage over the API?

Since 2.0.2 there is a route for it. GET /update shows the installed version and the version WordPress.org offers. POST /update {"install": true} installs the offered version straight away, the way the Update button does: the plugin stays active and the site shows its maintenance notice for a few seconds. POST /update {"check": true} asks WordPress to look for a new version now. The route runs WordPress’s own update code, takes the build from WordPress.org and nowhere else, and cannot be given a file or an address. Installing needs the right to update plugins (an administrator, or a network administrator on a multisite), and a site that forbids file changes is left alone. An API key with the default scope cannot use this route: it takes a key with the wider scope, an application password or a browser session. The plugin does not change how WordPress updates plugins; automatic updates stay as you set them on the Plugins screen.

How do redirects work?

POST /redirect with from (a path on your site), to (a path or a full URL) and type (301, 302, 307 or 410). If Rank Math’s Redirections module is on, the rule is saved through Rank Math and shows in its own list. Otherwise the plugin stores the rule and serves it. Redirects need an administrator, and Rank Math’s own Redirections capability when the rule goes through Rank Math.

What is left behind if I delete the plugin?

Deleting the plugin removes its API key hashes, so no way of signing in is left behind. Stored schema, redirect rules, the merchant policy and content backups are kept in the database, because a site may still want them, and they come back into use if the plugin is installed again.

Does it work with WooCommerce products?

Yes. Meta is registered on every public post type, so products, custom post types, etc. are all covered. Product category and product tag archives (and any other custom taxonomy) are also editable via the term endpoints (v1.3.0+).

Can I edit category / tag archive SEO meta in bulk?

Yes (v1.3.0+). Pass ?include_terms=1 to /export to pull category, tag, and any custom-taxonomy archive rows alongside posts. Post /bulk items with kind: "term" and the taxonomy slug to update them. Yoast term meta is stored in the wpseo_taxonomy_meta option; Rank Math uses standard term meta. Both are handled transparently, you pass plugin-neutral aliases (title, description, og_title, etc.) and the plugin writes to the right place.

Why am I getting 401 Unauthorized on localhost?

WordPress disables Application Passwords on non-HTTPS sites by default. For local development only, drop a small mu-plugin into wp-content/mu-plugins/ with add_filter('wp_is_application_passwords_available', '__return_true');.

Why does `curl` work but my Python/Node script gets 403 Forbidden?

The host’s web-application firewall (Apache mod_security, Wordfence, Solid Security, Cloudflare WAF) is rejecting the request based on User-Agent before WordPress sees it. Default UAs like python-requests/2.x and Node’s bare node-fetch are on most WAF blocklists. Send a regular browser User-Agent header on every REST call (e.g. Mozilla/5.0 ... Chrome/124.0 Safari/537.36) and the 403 disappears. The bundled MCP server (v1.4.2+) already does this automatically, override with MCP_USER_AGENT if your host has a stricter rule. Symptom signature: curl -u 'user:pass' https://site/wp-json/seo-meta-bridge/v1/status returns 200, but the same call from Python/Node with the same credentials returns 403 with an Apache HTML error page (not a WP JSON error).

Can the plugin be used to write arbitrary postmeta?

No. Only meta keys belonging to the active SEO plugin (Yoast or Rank Math) are accepted, plus the plugin’s own schema and schema_mode fields and, since 1.9.0, post_title. Other keys are rejected with unknown_or_disallowed_key.

Changelog

2.6.2

  • Divi Theme Builder layouts (the global header, footer and body templates) can now be read and edited through /content, /content/replace, /content/backups and /content/revert, with a backup before every write. Before, they answered “post not found”.
  • A layout shows on many pages, so it takes more than edit rights: the account must be able to change the theme, edit other people’s posts, post unfiltered HTML and use the Divi Theme Builder (Divi > Role Editor). Anyone else gets “post not found”. GET /content marks a layout with site_wide: true, and Divi’s cache for it is cleared after each change when Divi offers that.
  • A layout write or revert is refused, with nothing changed, when saving would change the layout beyond the edit: the content filter stripping or rewriting markup already in it (a script in a code module, an ampersand in a link) unless the site allows unfiltered HTML for the request, or WordPress’s own save fixes (entities, tag balancing), which nothing turns off. Edit such a layout in the Divi Theme Builder.
  • base_hash is now checked on post content writes too, as documented. Post content writes and reverts report the hash of what WordPress actually stored, with stored_differs: true when it adjusted the content on save.
  • /status reports supports_divi_layouts.

2.6.1

  • One-click sign-in keeps working when an AI app refreshes several times at once after its hourly token expiry, or retries after a lost answer: within 30 seconds they all get the same answer.
  • Demoting an administrator ends their connected apps; reusing a sign-in code ends the connection it made; a failed save asks the app to try again instead of handing out tokens that never work.
  • The approval page says when the site already holds the most connected apps, and when the site needs HTTPS. Error pages answer with status 400.
  • Stricter localhost return addresses, connections survive a permalink change, old end markers are cleared after a month, and uninstall removes every sign-in leftover.

2.6.0

  • One-click sign-in for claude.ai, the Claude desktop app and ChatGPT. Add the MCP address as a custom connector, click Allow on your site, done. No key to copy.
  • OAuth 2.1: authorization code with PKCE (S256), dynamic client registration, discovery documents, rotating refresh tokens. Only administrators approve, in their browser, and only Claude, ChatGPT, Cursor and apps on the same computer can ask. A password change ends that account’s connections.
  • A connection works on the MCP server only. Settings > AI CLI Connector lists connected apps with a Disconnect button and can switch the feature off.
  • /status reports supports_oauth.

2.5.0

  • MCP server built in. The plugin answers MCP requests at /wp-json/seo-meta-bridge/v1/mcp, so Claude Code, Cursor, VS Code and other MCP clients connect with one address and a key, with nothing to install. 15 tools, each running one of the plugin’s own routes with the same permission checks.
  • API keys are also accepted as Authorization: Bearer, the form MCP clients send.
  • New setup screen: a checklist with ready setups for Claude Code, for Cursor and VS Code, and a message for other AI tools; a check that the site can be reached; when and from which client your AI last used the site; and short Undo and Disconnect sections.
  • /status reports supports_mcp and mcp_url.

2.4.0

  • Easier setup. The key screen moved from Tools > Agent API Keys to Settings > AI CLI Connector, and the old address forwards to it.
  • A Settings link next to the plugin on the Plugins screen.
  • After activation, a dismissible notice on the Dashboard and Plugins screen offers Set up now. It goes away once a key is made or it is dismissed, and an update never shows it.
  • The setup screen walks through three steps, and a new key is shown together with the site address and username behind one Copy connection details button. The curl examples moved under “For developers”.

2.3.2

  • Live SEO panel: no price, offer or product checks on pages that are not WooCommerce products, so a service page is no longer told to add a price nobody set.
  • Live SEO panel: fewer busywork warnings. A short title or description and a page without structured data are notes now; the Social row warns only without a share image; nested organizers and sponsors are not checked as the business.
  • Live SEO panel: the canonical check ignores tracking parameters (utm_*, gclid, fbclid and similar) and parameter order.
  • GET /status reports merchant_enrichment.enabled false when WooCommerce is not active.

2.3.1

  • Fix: POST /update with check answered “up to date” without asking WordPress.org when the stored time of the last update check was in the future. The “checked less than a minute ago” shortcut took any future time as recent, so the check was skipped until the server’s clock caught up (seen on a live site, about an hour ahead). Only a last check between 0 and 60 seconds ago counts now; a future one is treated as stale and the check runs. If WordPress.org does not answer, what WordPress knew is put back and the route answers 503, as before; a last check stamped in the future is put back as 0, so WordPress’s own scheduled checks are not held back either.
  • GET /update also reports server_time and checked_in_future, so a stored check time ahead of the server’s clock is visible.

2.3.0

  • New: live SEO panel. Signed-in users who may edit the page in view get an “SEO” item in the admin bar with a count of problems. It opens a panel over the page with four tabs. Overview: the search result and every tag read from the rendered page (title, description, robots including an X-Robots-Tag header, canonical, H1, social tags, images, links), each judged. Edit: SEO title, description and focus keyword with length checks, saved into Yoast SEO or Rank Math. Schema: every JSON-LD block with the plugin that printed it, missing required and recommended properties, duplicate Organization or WebSite nodes, broken JSON, links to an @id that is not on the page; custom schema can be added or set to replace the page’s schema. History: the SEO changes on this post.
  • New: change history. Every change to a post’s SEO title, description, keyword, canonical, robots, social fields or custom schema is kept on the post (the last 30), with the time, the user and where it came from: the editor, an API key, the REST API, WP-CLI or the live panel.
  • New: “Live on the site” in the block editor’s document sidebar and as a side box in the classic editor: what the published page sends, read again after every save, with the last changes.
  • New route POST /live, for the panel only. It needs a signed-in browser session with a REST nonce (keys and application passwords use /bulk), the right to edit that post or term, and each field passes the same checks as in /bulk, including the SEO plugin’s own role rules for its settings groups.

2.2.0

  • New: Elementor choice settings in /content. Dropdown, choose and switch controls on a widget’s Content tab, including the fields of repeater rows (for example a Post Info row’s type, taxonomy and link), are listed with kind: choice and their options. A write is accepted only when the value is one of those options. Heading levels stay kind: tag. Needs Elementor running, because the options come from its registered controls.
  • New: add an Elementor element by copying an existing one. POST /content {"inserts": [{"ref": "<element id>", "position": "after", "copy_from": {"post_id": 123, "el": "<element id>"}}]} clones an element from the same page or another Elementor page you can edit, with new element and row ids and without its CSS ID, before, after or inside ref. Elements only go where Elementor accepts them (a section placed in a column becomes an inner section). Up to 20 per request, applied before updates[], with the usual dry run, base_hash, backup and revert. New atomic=1 writes nothing when any update or insert in the request fails.
  • Security: the copy source passes the same checks as editing it (edit rights, the builder gate and Elementor’s own permission). Unless the site allows raw HTML (the bulkseme_allow_unfiltered_html filter) and the request sends unsafe_html=1, a copy, including an administrator’s, keeps only the standard element keys (no local styles), its text is cleaned like a direct write (the result says sanitized: true when anything was cleaned or left out), and they cannot copy HTML, shortcode, template, code or WordPress widgets, any widget with a code control, Custom CSS, custom attributes, dynamic tags, or any other setting that holds markup or a script link; nor set a choice to a raw output value. Size limits per copy, per request and per page (8 MB, 24 levels) stop runaway growth.
  • /status reports supports_choice_settings and supports_inserts.

2.1.0

  • New: add a row to an Advanced Custom Fields repeater or flexible content field. Send an update {"op": "add_row", "el": "<field>", "position": 2, "layout": "<layout>", "values": {"<sub field>": "text"}} in POST /content. GET /content lists every repeater and flexible content field under containers, with its row count and the sub fields a new row may fill. The rows below the insert move down with every value they hold, images and nested rows included. Text is cleaned like any other write, the field’s maximum row count holds, and a stored value that belongs to another field, or that names a PHP class, stops the insert before anything is written. The backup holds the whole field, so /content/revert puts the rows back exactly.
  • New: Advanced Custom Fields on taxonomy terms. With kind=term, GET /content lists the term’s ACF text fields and rows under acf, and POST /content with updates[] edits them and adds rows, with the same backup and revert as on posts. The term description is left alone.
  • Fix: a write to text kept in post meta (ACF, Beaver Builder, SiteOrigin, Accordions, and Elementor when its own resync is not available) bumped the modified date by saving the post again. For a user without unfiltered_html that ran the stored title, content and excerpt through kses and could rewrite markup the write never aimed at, such as the trailing semicolon of a style attribute. Only the date changes now.
  • Security hardening: kind=term no longer reaches a taxonomy that has neither an edit screen nor a REST route (such as post formats), the same reach WordPress itself gives. Present since 2.0.0.
  • Security hardening: on a network, the fallback that reads ACF field definitions from stored meta when no field group matches is for super admins only.
  • New filter bulkseme_user_can_edit_term_acf keeps a role out of a term’s ACF fields, as bulkseme_user_can_edit_with_builder does for posts.
  • /status reports supports_acf_rows and supports_term_acf.

2.0.2

  • Security: /import read an uploaded CSV file line by line, so one quoted cell that held line breaks was read as several rows. A post title can hold line breaks and /export wrote it as it was, so an Author could plant rows in a title, and they were applied with the rights of whoever re-imported that export: SEO fields of posts the Author may not edit, and the site-wide title settings. The file is now read as a whole, the way RFC 4180 defines it, and line breaks and other control characters are taken out of every exported cell. If you read an export in a script, use a CSV reader over the whole text, never line by line. Present in every earlier version. Found in an independent review; not known to have been used.
  • Security: stored JSON-LD schema is no longer printed on a password-protected post until the visitor has entered the password.
  • Security hardening: a key with the default scope signs in on this plugin’s own routes by name, and no longer on everything under its REST namespace, so it cannot reach a route another plugin registers there.
  • Fix: /export ended with a PHP error when post_type, status or taxonomy was sent as a list (?status[]=draft). Lists are accepted now. A list sent where one value belongs is refused cleanly, and a list sent as an id is no longer read as post 1.
  • Fix: a URL field (canonical, Open Graph image, Twitter image) holding bytes that are not valid UTF-8, written through WordPress’s own REST route, was answered with a 500. Such bytes no longer reach the database.
  • Hardening: WPBakery design rules rebuilt with rebuild_wpb_css go through the filter WordPress uses for inline styles. HTML written to a SiteOrigin widget or Beaver Builder module that the plugin has no field map for has the quotes in its text stored as entities.
  • /import of a file: 10 MB limit, a file that ends inside a quoted cell is refused, and rows whose cell count does not match the header are counted in skipped_rows.
  • New: GET and POST /update. See which version is installed and which one WordPress.org offers, and install the offered version without opening wp-admin. It runs WordPress’s own update code, so the build can only come from WordPress.org; installing needs the right to update plugins, a site that forbids file changes is left alone, and a key with the default scope cannot use the route.
  • Security hardening: a key that is revoked here is also taken out of the copy a companion build keeps of the keys it handed over, whether or not that build is active at the time.
  • MCP server 1.7.2: list_posts lists only post types the site exposes, every tool argument that reaches a request path is checked, requests time out after 60 seconds, an answer over 20 MB is refused, and the dependency lock file is refreshed.
  • The GitHub repository moved to github.com/puneetindersingh/bulk-ai-cli-connector.

2.0.1

  • Security hardening: /clone now checks a stored meta value in exactly the form WordPress reads it back. A value planted straight into the database with a NUL byte in front of a serialised object could get past the check that keeps /clone from recreating objects of other classes. Using it needed a row written outside WordPress and a caller with the unfiltered_html capability. Found in an independent review; not known to have been used.

2.0.0

  • Renamed from “Bulk SEO Meta Editor for AI Agents” to “Bulk AI CLI Connector”. The plugin address, folder, REST routes and stored data are unchanged.
  • New: page content. GET /content and POST /content read and edit the visible text of a page for Elementor (classic and Elementor 4 atomic widgets, heading level, link URLs and per-element custom CSS), Beaver Builder, SiteOrigin Page Builder (core text widgets and the SiteOrigin Widgets Bundle: editor, headline, button, call to action, accordion, tabs, hero, features, testimonials and image), the Accordions plugin and Advanced Custom Fields text fields, and the post content of Gutenberg, classic, Divi and WPBakery pages. POST /content/replace runs a literal find and replace. Add kind=term&taxonomy=<slug> for a term description, which keeps headings, lists and tables.
  • New: a backup before every content write, with GET /content/backups and POST /content/revert. Backups are kept in a hidden post type that nobody can read or edit outside these two routes, the newest ten per item, and a backup can only be restored onto the item it came from.
  • New: dry_run and base_hash on content writes.
  • New: POST /clone copies a page or post into a new draft with its template, design meta, featured image and terms. SEO fields are not copied.
  • New: GET, POST and DELETE /redirect. Rules go through Rank Math’s Redirections when that module is on, otherwise the plugin stores and serves them.
  • New: GET, POST and DELETE /merchant-policy. Adds return policy, shipping details, brand and a price start date to WooCommerce Product schema that the SEO plugin left without them. Opt out with add_filter('bulkseme_merchant_enrichment', '__return_false').
  • New: API keys. Made on Tools > Agent API Keys, shown once, stored as SHA-256 hashes, REST only, HTTPS only. Send a key in the X-Bulkseme-Key header or as the password of HTTP Basic auth. A key is accepted on this plugin’s own routes only, unless it is made with the wider scope, which covers every REST route. A key is tied to the account that made it and is removed when that account is deleted. POST /register-key stores the hash of a key from an administrator’s own browser session.
  • Security: content routes check edit_post or edit_term per item and follow the page builder’s own access rules (Elementor’s excluded roles, Beaver Builder’s Builder Access). Every value is cleaned with wp_kses_post, administrators included. Storing unfiltered HTML needs the bulkseme_allow_unfiltered_html filter, the unfiltered_html capability and unsafe_html: 1 in the request.
  • Security: redirects and merchant policy need manage_options. Redirects written through Rank Math also need its rank_math_redirections capability.
  • Security: posts of a type that has no edit screen and no REST route of its own are now out of reach of every route, including /bulk, /export and /schema. Before this, a user could rename such a record through post_title in /bulk if WordPress counted them as its author, and that included a privacy request they had filed themselves, whose title is the address the request acts on. Recommended update for every site.
  • Security: stored JSON-LD is limited to 128 KB and 5,000 values per post or term (schema_too_large), because it is read on every view of the page. Filters: bulkseme_schema_max_bytes, bulkseme_schema_max_values.
  • Security: on a site with neither Yoast SEO nor Rank Math, storing schema now needs edit_others_posts (Editor and above), in line with Yoast SEO’s own default for schema settings. Filter: bulkseme_user_can_edit_field_group.
  • Fix: titles and descriptions that use replacement variables are stored intact. WordPress’s general text cleaner removed anything that looked like a percent-encoded byte, so %%category%% was stored as %tegory%% and Rank Math’s %date% as te%. The active SEO plugin’s own text cleaner is now used. Post titles keep their percent signs too.
  • Fix: a merchant policy number that is not finite (for example 1e400) is refused. It could not be printed as JSON and emptied the product schema block.
  • /status reports the new capabilities: supports_content, supports_redirects, supports_api_key, supports_clone, builders and more. Page builder version numbers, the redirect engine and the merchant policy summary are only shown to administrators. Filter: bulkseme_status.
  • MCP server 1.7.1: post ids and post type names are checked before they are used in a request path.
  • The plugin is now split into files under includes/. No setting, stored value or existing route changed name.
  • Deleting the plugin now removes its API key hashes, on every site of a network.

1.2.0 to 1.9.1

  • Earlier releases: Yoast and Rank Math field mapping, bulk and CSV routes, term and archive SEO, image alt text, custom schema, post titles and sitemap flushing. Full notes are in the GitHub repository’s release history.