![\"Respira](\"https://www.respira.press/images/mcp-hero.png\")
\n \n\n \n \n \n
\n 283 MCP tools, 387 with the WooCommerce add-on. 19 page builders and the native Site Editor. MCP 2026-07-28 with legacy-client support.
\n Element-level editing, full page creation, HTML-to-builder conversion with rendering fidelity validation, per-site tool governance, storefront design intelligence, stock images, bulk operations — all through one MCP server.\n
\n ![\"npm](\"https://img.shields.io/npm/v/@respira/wordpress-mcp-server.svg?style=flat-square&color=10b981\")
\n ![\"npm](\"https://img.shields.io/npm/dm/@respira/wordpress-mcp-server.svg?style=flat-square\")
\n ![\"366](\"https://img.shields.io/badge/tools-387-10b981?style=flat-square\"){kind=icon}
![\"19](\"https://img.shields.io/badge/builders-19-10b981?style=flat-square\"){kind=icon}
![\"TypeScript\"](\"https://img.shields.io/badge/TypeScript-100%25-blue?style=flat-square&logo=typescript&logoColor=white\"){kind=icon}
\n Website •\n Docs •\n Plugin •\n Support •\n WooCommerce Add-on\n
\n\n---\n\n## Reviewed menus and redirects\n\nOn plugin 9.2.13+, a Publish only key can propose menu and redirect changes without writing them directly:\n\n| Tool | Purpose |\n| --- | --- |\n| `wordpress_read_change_target` | Read a complete target state and version. |\n| `wordpress_create_pending_change` | Save exact before/after for review. |\n| `wordpress_get_pending_change` | Read one proposal and its durable receipt. |\n| `wordpress_list_pending_changes` | Filter and page through site-local changes. |\n| `wordpress_approve_pending_change` | Apply only the reviewed version and unchanged live target. |\n| `wordpress_reject_pending_change` | Reject without changing the public site. |\n| `wordpress_undo_pending_change` | Create a reverse proposal requiring its own approval. |\n| `wordpress_expire_pending_changes` | Mark stale proposals expired, retaining history. |\n| `wordpress_propose_status_change` | Propose unpublishing a page or post to draft, or republishing it, in one call (plugin 9.2.20+). |\n\nRead `published`, `complete`, `warnings`, and `applied` in the approval receipt. A successful HTTP status alone is not full success. Never retry an incomplete apply. New items use `new:your-reference`; receipts map those references to assigned ids. Undo recreates deleted entries with new ids. Menu location reassignment is outside v1. Respira, Redirection and (plugin 9.2.27+) SmartCrawl are writable; other redirect managers remain read-only. A proposal can add up to 50 redirects at once, which makes it the reviewed way to import a batch.\n\nPages and posts work the same way (target kind `post_status`, plugin 9.2.20+). A Publish only key can propose taking back a page it created once a person published it: `wordpress_propose_status_change` with `status: \"draft\"`. The page keeps its slug, its URL returns 404 while it is a draft, and no redirect is created. `wordpress_undo_pending_change` on the applied change proposes republishing it at the same URL. A person approves each step; a Publish only key never approves a status change itself.\n\n## What Makes Respira Different\n\nOther WordPress MCP servers wrap the REST API. They can create posts and pages, but **they can't touch your page builder content**.\n\nRespira includes a WordPress plugin that gives AI **native access** to 19 page builders, plus element-level precision, full page creation from structure, HTML-to-builder conversion, storefront design intelligence, stock image search, and bulk operations across hundreds of pages.\n\n### New in 8.6: full Yoast SEO, Avada, approvals bound to what a person saw\n\n- **Full Yoast SEO support (plugin 9.2.10+).** The `seo` field on `respira_update_page`, `respira_update_post` and `respira_update_custom_post` writes every Yoast field: title, meta description, focus keyphrase, canonical, breadcrumb title, robots, Open Graph and X cards, cornerstone, schema types and primary category, through Yoast's own functions so its index and sitemaps stay in step, and each field is read back. On Yoast sites four tools appear: `respira_get_yoast_seo` (the rendered head, SEO and readability scores, internal links, sitemap inclusion), `respira_update_yoast_term_seo`, `respira_get_yoast_search_appearance` and `respira_update_yoast_search_appearance`. Rank Math, All in One SEO 4, SEOPress and SmartCrawl are written through the same `seo` field, and a site running an SEO plugin Respira cannot write to gets a warning naming it instead of a silent success.\n- **Publishing bound to the version a person approved (plugin 9.2.10+).** `respira_approve_duplicate` takes `expected_version`: if the draft changed after someone looked at it, nothing is published, and `force` does not override it. Approval carries the SEO fields over, reads them back on the live page, and snapshots include them, so undo restores them exactly.\n- **A Publish only key.** Made in the site's wp-admin under Respira, Settings, API keys: drafting on duplicates, previews, SEO fields and version-bound approval, nothing else. The server lists only the tools a key may call.\n- **Avada is builder number 19 (plugin 9.2.6+).** Avada Builder pages are read and written as Avada's own shortcodes, with 11 tools for its design system, Layouts and Library on Avada sites.\n- **Inserts that leave the rest of the page alone.** `respira_inject_builder_content` takes `target_path`, `position`, `after_element_id`, `before_element_id`, `chunk_id` and `session_id`: on Divi only the new content is written, so global modules, Divi 5.15 Components and third-party modules stay exactly as stored, and big pages can be built in small chunks that are never added twice.\n- **Native block-library blocks (plugin 9.2.7+).** Kadence Blocks, Spectra, GenerateBlocks and GreenShift pages get their own blocks when your AI names the library or the page already uses it.\n- **A busy site is retried, not reported.** A site runs at most six Respira calls at once; the server waits what the site asks and sends the call again. PageSpeed audits and rendered design checks run in the background and the server waits for their result.\n- **Also in 8.6:** Murmur notes arrive with their photos and pinned screenshots, `respira_request_upload` gives a one-time upload link for a file, publishing a change is asked once in the chat, Elementor 4.3 nesting guidance, and more WooCommerce tools.\n\n### New in 8.5: YOOtheme Pro, and redirects that look before they leap\n\n- **YOOtheme Pro is builder number 18.** Pages are read and written as YOOtheme layouts (sections, rows, columns and YOOtheme's own elements) and saved through YOOtheme's own builder, so they open in the customizer like a page built there. Your AI gets every element's fields from the running theme, custom elements included. Four template tools appear on YOOtheme sites for the layouts YOOtheme renders on single posts, the blog index, archives, search and the 404 page; a new template starts disabled, so nothing changes until you have looked. Plugin 9.2.0+.\n- **Redirects with guardrails.** `respira_list_redirects`, `respira_create_redirect`, `respira_update_redirect` and `respira_delete_redirect` write through Redirection when it is active and through Respira's own small list otherwise; Rank Math, Yoast SEO Premium, Safe Redirect Manager and Simple 301 Redirects are read, never written. WordPress's own addresses are refused, loops through any redirect plugin are refused, a 301 to a dead page is refused unless you say why, and the home page or a catch-all pattern waits for a person's yes. Every change keeps a snapshot. Plugin 9.2.0+.\n\n### New in 8.3: Design Direction, the site's design system as an artifact\n\n- **One saved design direction per site.** Identity, DTCG-subset tokens, plain-language guidance including the don'ts, and waivers, saved on the site with revisions. Agents read it before building anything; activation refuses a direction that is missing its core roles and names exactly what is absent.\n- **Any token format in, strict DTCG out.** Import DTCG 2025.10, Tokens Studio exports, Tailwind configs, or pasted `:root` CSS into a draft (dry-run by default), apply into the builder's NATIVE token store with a per-builder report, export as strict DTCG 2025.10 with builder mapping metadata under `$extensions[\"press.respira\"]`.\n- **A design check that can say \"not done yet\".** `wordpress_check_design` runs deterministic slop-and-drift rules against draft content or a saved post: off-palette colors and fonts, filler copy, placeholder names, AI-default palettes, and probes generated from the direction's own don'ts. With `rendered: true` (plugin 8.6.20+) it renders the real page at desktop and mobile widths: contrast, type-scale coherence, three equal cards, hero filling the viewport, section repetition, spacing rhythm, broken layout, and off-palette colors as actually painted, with screenshots attached. Anything it cannot verify is listed as `unchecked` instead of silently skipped.\n- Build skills run the check before treating any page as done.\n\n### New in 8.2: Resonance, a persistent memory for every site\n\n- **Your AI forgets your site every session. Resonance remembers.** `wordpress_remember`, `wordpress_forget`, and `wordpress_list_memory` maintain a capped, per-site store of conventions, preferences, and lessons that rides into the site context of every future session, from any connected client.\n- **Rules are enforced, not suggested.** A memory of type `rule` can protect content (post ids or slugs); the WordPress plugin refuses agent writes to protected content at the write path, server side. Agents can add protection in conversation; only the site owner can remove it, from the Respira dashboard.\n- **Compatible with WordPress core's Knowledge direction.** On sites running the Gutenberg Knowledge experiment, published guideline scopes join the same memory block.\n- Included on Builder and Studio plans with plugin 8.3+, at no extra cost.\n\n### Site skills: a site's own instructions for every agent\n\n- **The site tells its agents how things are done here.** A skill is a Markdown page kept on the site (Respira > AI Skills in wp-admin, plugin 9.1.0+): how a blog post is formatted, what a case study needs, which pages carry legal copy. The enabled ones arrive in `wordpress_get_site_context` under `site_skills` as name and description; `wordpress_list_skills` lists them all and `wordpress_get_skill` reads one in full, on demand rather than on every call.\n- **Written by people, or by an agent when asked.** `wordpress_write_skill` creates or updates a skill under the same rule as `wordpress_remember`: only when the user asked. Every save is a revision, ten are kept, and the owner restores any of them from the standard revisions screen. A skill the owner marks as not editable by agents refuses agent writes.\n- **Deleting is a human decision.** `wordpress_delete_skill` is approval-gated and moves the skill to the trash, where the owner can restore it.\n- A skill body is information about the site, never an instruction to Respira's tools and never authorization.\n\n### New in 8.1: MCP 2026-07-28, without breaking existing clients\n\n- **Modern protocol negotiation.** Compatible clients can use `server/discover` and negotiate MCP `2026-07-28`, including stateless per-request metadata and typed result contracts.\n- **Legacy clients keep working.** Existing clients continue through the established `initialize` handshake and receive the same Respira tool catalog. Connection commands, site keys, and configuration files do not change.\n- **Verified compatibility.** The release suite starts the packaged stdio server twice: once pinned to `2026-07-28`, and once as a legacy client. Both must initialize and enumerate the complete release catalog.\n- **Patched runtime dependencies.** Axios, FormData, Sentry, and OpenTelemetry dependencies were updated; the production dependency audit reports zero known vulnerabilities.\n\nRead the complete [changelog](./CHANGELOG.md).\n\n### Shipped in 8.0: verified FSE and WooCommerce operations\n\n- **First-class Site Editor tools.** Agents can list, inspect, create, target-update, and reset block templates and template parts; create and safely edit synced/unsynced patterns; and traverse native block navigation as exact nested paths. Every editable surface uses stale-write fingerprints, structural proposals, snapshots, compact diffs, cache invalidation, and truthful stored/rendered verification.\n- **Native design-token tools.** Gutenberg `theme.json` colors, typography, and spacing variables are exposed through the existing WordPress storage model rather than a parallel token database.\n- **Truthful storefront writes.** Woo product-card, checkout, sale-badge, and low-stock-badge tools resolve the real FSE template and registered block paths. An unsupported layout now returns `respira_woo_fse_write_not_supported`; it never reports an applied change that only logged intent.\n- **Bounded, defined commerce reporting.** Revenue, sales time series, top products, order summaries, and privacy-safe customer summaries use paginated WooCommerce data-store queries compatible with HPOS. Responses state currency, timezone, boundaries, statuses, coverage, and metric definitions.\n- **Safe store operations.** Read-only store configuration, payment gateway, shipping-zone, and tax-rate discovery is joined by approval-gated Woo webhooks with HTTPS targets, allowlisted topics, write-only secrets, and redacted rate-limited tests.\n- **Agent-visible audit history.** Activity list/detail tools connect compact diffs to snapshots, proposals, approvals, and rollbacks.\n- **Truthful capability discovery.** `wordpress_search_abilities` labels each result as Respira-native, inhaled, unavailable, or replaced by a safer audited Respira workflow, and returns the exact next action.\n- **Explicit feed recovery.** `woocommerce_get_feed_status` stays read-only while `woocommerce_repair_feed_scheduler` can clear only a confirmed-stale Respira lock, resume the idempotent build, verify the repair, and preserve Action Scheduler history.\n- **Safer plugin activation.** Dependency/version preflight, a signed fresh-process activation probe, REST/frontend boot checks, and narrowly scoped rollback prevent a broken plugin from taking the controlling request down with it.\n- **Generated release integrity.** A public [tool catalog](./TOOL_CATALOG.md) and machine-readable `tool-capabilities.json` are generated from the declarations used in release CI. Missing native or documentation coverage requires an explicit, reviewable exclusion.\n\n### New in v7.6.2: clearer failures and trustworthy quality signals\n\n- **PageSpeed failures keep their real machine-readable reason.** Rate limits, upstream outages, proxy HTML responses, and other structured WordPress errors no longer collapse into a generic failure, so agents can distinguish “retry later” from a Respira defect.\n- **PageSpeed tools reject incomplete requests before touching WordPress.** Both PageSpeed schemas require either `page_id` or `url`, preventing avoidable calls that could only fail.\n- **Every transport attempt has its own identity.** The logical call id remains stable for replay-safe writes, while `attempt_id` preserves each network attempt so a retry can never overwrite the evidence from the first timeout.\n- **Privacy remains metadata-only.** Retry detection uses a process-secret target hash. Respira never sends prompts, raw arguments, tool results, or WordPress content.\n- **The admin quality dashboard now separates operational health from task completion.** Expected approval/safety blocks, caller mistakes, customer setup, upstream services, product errors, unknown failures, and incomplete attempts are measured independently. Alerts use comparable site cohorts and require failures across multiple sites, reducing false alarms caused by one unusual customer setup.\n- **The npm package no longer publishes private repository metadata.** Support and product links point to respira.press.\n\n### Reliability and builder additions since v7.5.2\n\n- **v7.5.3:** Windows media paths and `file://` uploads work correctly; `delete_media` approval tokens now reach the confirmation call.\n- **v7.5.4:** `tools/list` shares one cached, two-second-bounded context fetch instead of making three sequential WordPress checks, preventing slow-site handshake disconnects.\n- **v7.5.5:** subdirectory WordPress installs on the same hostname receive distinct site identities; custom Elementor widget guidance now uses live control schemas and targeted stable-ID updates.\n- **v7.5.6:** a successful `?rest_route=` diagnostic fallback becomes sticky for the rest of the MCP session.\n- **v7.6.0:** WPML translation tools and Divi Theme Builder header, footer, and body-template tools.\n- **v7.6.1:** `wordpress_build_mega_menu` creates a complete native Divi 4 or Divi 5 mega-menu hierarchy in one call.\n\n### Reconnect-safe writes and drop diagnostics (v7.5.2)\n\n- **Writes carry a stable idempotency key.** The connector automatically retries transient reads. It retries a write only when WordPress explicitly confirms replay support; an ambiguous timeout/reset is reported as `respira_write_outcome_unknown` so the agent verifies the target before trying again.\n- **Telemetry queues and retries independently of tool execution.** A dashboard or network interruption cannot block a WordPress call, and failed metadata batches are retained with bounded exponential backoff.\n- **Local API keys stay local.** With Respira for WordPress 7.5.46+, a local `respira_*` key is exchanged inside WordPress for a revocable, site-scoped telemetry bearer. The central service never receives that local key.\n- **The quality dashboard can separate exact retries from editing sweeps.** It records stable session/call identity, transport, incomplete calls, connector/plugin versions, total vs tool time, and privacy-safe stage durations. Target comparison uses a secret HMAC; arguments, results, prompts, and WordPress content are never sent.\n- **Native URL-connected clients gain timeout-safe writes.** Plugin 7.5.46 adds `_respira_async: true`, `respira_get_job_status`, stable `Mcp-Session-Id` responses, replay-safe `_respira_call_id`, and a native `respira_diagnose_connection` tool. The native endpoint is stateless HTTP—independent POST requests, not a persistent “write channel.”\n\n### Also shipped between v7.1 and v7.5.1\n\n- **v7.2:** cascade session rollback; dropped-styling/no-op guidance; page-parent schemas; Oxygen 6 build guidance; completed live-edit confirmation parameters; bounded heavy-write/docs timeouts; and route-aware capability negotiation that hides tools an older plugin cannot execute.\n- **v7.3:** the WooCommerce add-on expanded from 21 to 56 tools across catalog, pricing, stock, storefront, brands, variations/attributes, coupons, customers, refunds, and AI-readiness.\n- **v7.4:** Agent-Ready Commerce expanded WooCommerce to 79 tools with feeds + store `llms.txt`, readiness fixes, signed cart links/attribution, and Subscriptions, Bookings, and Memberships writes. v7.4.1 then made newly connected dashboard sites self-heal into a running connector session.\n- **v7.5.0:** eight STAGGS configurator tools brought the WooCommerce total to 87.\n- **v7.5.1:** fixed normalized ACF tool routing and preserved dashboard-token 401 machine codes.\n\n### New in v7.1 \"Roots\"\n\nThe biggest release since launch.\n\n- **Four new page builders.** Spectra, Kadence Blocks, and GenerateBlocks join with full read + write; SeedProd ships read + audit. Supported builders go from 12 to 16.\n- **Make a page responsive in one call** on Bricks, Elementor, Divi 4, Beaver Builder, and WPBakery.\n- **Playbooks.** Teach an agent a multi-step workflow once and rerun it as a single callable WordPress Ability. Five tools (create / list / get / update / delete), with static cycle detection and a runtime depth cap.\n- **Agent-creatable Custom Post Types, taxonomies, and ACF field groups** (twelve tools), stored in options and registered on `init`. No PHP files, server-side slug allowlist, conservative deletes that report orphans.\n- **Native in-WP MCP endpoint.** Point a client straight at `your-site/wp-json/respira/v1/mcp` (Streamable HTTP JSON-RPC; `?rest_route=` fallback supported) with an API key, no npx process required. One-click connect-by-link mints and installs the key for you.\n- **System certificate-store trust.** The connector trusts the OS certificate store at startup (verification stays on) with a bundled Mozilla CA fallback, so valid public certs stop being rejected on Claude Desktop's bundled Node.\n- **Bricks design-token import + section presets**, the **Elementor 4 atomic write path** (partial: 20 atomic element types with `$$type` envelopes, per-breakpoint + per-state styling, and a clear 422 on unsupported types instead of a silent drop), and **`respira_generate_activity_report`** for client-ready reports.\n\n### New in v6.17.0: an MCP that knows how to behave\n\nThree additions plus the agent persona that ties them together.\n\n- **SOUL.md, loaded into every handshake.** Identity, voice, values, and rules of engagement an AI client reads before its first tool call. No \"we.\" Privacy hard line. Builder-native edits only. Try yourself before asking the customer. Mirrored at [respira.press/soul.md](https://www.respira.press/soul.md) for outside-MCP integrations (Cursor system prompts, Claude Projects, GPT custom instructions).\n- **`respira_search_docs`**, full-text search across the Respira documentation. The agent calls this before offering to file a bug because most \"bugs\" are documented known issues with workarounds. Public, unauthenticated, free to call.\n- **`respira_report_issue`**, files a structured bug report from inside the AI chat directly to the maintainer. Auto-attaches site URL, builder + version, MCP version, OS, last tool, optional `respira_diagnose_connection` snapshot. Privacy hard line: prompts, tool args, tool results, and WP content never enter the payload.\n- **`respira_diagnose_connection` probe timeout default 30s** (was 15s), new `probe_timeout_ms` arg clamped `[5000, 60000]`. Sites behind Cloudflare custom rules routinely cross 15s on first-hit HEADs while curl returns in under 4s; the old default produced spurious \"MCP timed out but the site is fine\" reports.\n- **Bootstrap stderr log includes the site list:** `respira-mcp vX.Y.Z ready · N sites: host1, host2, ...`, so you can tell at a glance whether the MCP came back after a Claude restart and which sites it loaded.\n\nThe triage rule, baked into the handshake instructions: tool error → apply hint + retry once → `respira_search_docs` → `respira_diagnose_connection` → ask the customer if they want to file a bug → `respira_report_issue`. Never auto-file.\n\n### v6.16.0: usage telemetry ships by default\n\nv6.16 introduced default per-tool telemetry so customers did not have to configure an OTEL bearer. As of v7.5.2 with plugin 7.5.46+, local WordPress API keys are never sent to the central ingest: the connector asks WordPress for a revocable, site-scoped telemetry token instead. Privacy remains metadata-only—no prompts, raw tool arguments, tool results, or WordPress content. `RESPIRA_USAGE_OPT_OUT=1` disables the entire stream.\n\n### v6.3 — MCP Protocol Compliance\n\nEvery successful tool response includes `structuredContent`, the raw JSON object, alongside the existing `content[0].text` (stringified JSON). Follows the MCP 2025-06-18 spec. Clients that understand `structuredContent` get direct programmatic access to tool results without parsing JSON from text. Older clients are unaffected; the `content` array is still there. Unknown tool names return a proper `CallToolResult` with `isError: true` and a hint to discover available tools, instead of throwing a protocol-level JSON-RPC error. Lets LLMs self-correct gracefully rather than hitting a hard protocol failure.\n\n### v6.0 \"Storefront\" — Context-Aware Tool Filtering\n\nThe MCP server automatically filters the list based on the site's detected builder and active plugins, so clients receive only the useful subset. Fail-open: if detection fails, the full list is returned.\n\n| Capability | Respira | Typical WordPress MCP servers |\n|---|---|---|\n| Page builder support | **19 builders** (incl. Spectra, Kadence, GenerateBlocks, Avada) | None, or a single builder |\n| Element-level find/update/move/remove | **Yes** | No |\n| Build full pages from structure | **Yes** | No |\n| Convert HTML to native builder | **Yes** | No |\n| Stock image search + sideload | **Yes** | Rare |\n| Bulk operations (100 pages/call) | **Yes** | No |\n| 27 widget shortcuts (add_heading, etc.) | **Yes** | No |\n| Duplicate-before-edit safety | **Yes** | Rare |\n| Snapshot rollback | **Yes**, builder-aware | Rare, post content only |\n| SEO / Core Web Vitals / AEO analysis | **Yes** | Rare |\n| WooCommerce (products, orders, inventory) | **Yes** (add-on) | Rare |\n| Tool governance (per-tool enable/disable) | **Yes** | Rare |\n\n---\n\n## Quick Start (3 Minutes)\n\n### Step 1: Install the WordPress Plugin\n\nDownload from [respira.press/plugin](https://respira.press/plugin) → upload to WordPress → activate → go to **Respira > API Keys** → generate a key.\n\n### Step 2: Configure Your AI Tool\n\nWe build amazing things.
\" } },\n { type: \"button\", settings: { text: \"Get Started\", link: \"/contact\" } }\n]})\n```\n\n### HTML-to-Builder Conversion\nConvert any HTML into native builder widgets — with CSS extraction, responsive mapping, and a fidelity report.\n\n```\nrespira_convert_html_to_builder({ html: \"\n ![\"Elementor\"](\"https://img.shields.io/badge/Elementor-Full_Intelligence-92003B?style=for-the-badge\"){kind=icon}
![\"Divi](\"https://img.shields.io/badge/Divi_5-Full_Intelligence-7c3aed?style=for-the-badge\"){kind=icon}
![\"Bricks\"](\"https://img.shields.io/badge/Bricks-Full_Intelligence-ef6723?style=for-the-badge\"){kind=icon}
![\"Flatsome\"](\"https://img.shields.io/badge/Flatsome-Full_Intelligence-3d9a8b?style=for-the-badge\"){kind=icon}