Skip to content

Compared with what exists

This summarizes the landscape research done on 17 October 2026, with primary sources read that day. [S] marks a claim seen only in a search snippet. Install counts and stars change daily.

Columns: 1L = one-line script install · AD = zero-config auto-discovery (JSON-LD/OG/feeds/sitemap/llms.txt) · FT = full-text search · MD = full article as Markdown (no tiny cap) · TOC = sections/TOC of current page · REL = related posts · TAX = taxonomies · SSG = works across SSGs · CMS = headless CMS adapters · WP = WordPress · PX = proxy mode for private tokens · SEC = read-only + untrusted hints + caps · TEST = tests/CI · NAV = desktop-agent bridge compat (MCP-B relay) · CUR = current API (document.modelContext)

Solution 1L AD FT MD TOC REL TAX SSG CMS WP PX SEC TEST CUR
Cloudflare WebMCP packs ✅ (edge toggle) ✗ ✗ ✗ ✗ ✗ ✗ any (CF-only) via own /mcp via own /mcp n/a ~ ? ✅
WebMCP Bridge (WP) ✅ plugin ✗ ~ (WP search) ~ (needs Mescio) ✗ ✗ ✅ ✗ ✗ ✅ ✗ ~ ? ?
Royal MCP (WP) plugin ✗ ~ ✗ ✗ ✗ ✅ ✗ ✗ ✅ server ~ (OAuth) ? server-side
Agentgate (WP) plugin ✗ ~ ✅ ✗ ✗ ✅ ✗ ✗ ✅ ✗ ✅ ? ✗ navigator
astro-webmcp 1 config line ~ (content collections) ✅ Pagefind/Orama ✗ (1.5k cap) ✗ ✗ ~ Astro only ✗ ✗ ✗ ✅ ? ✅
webmcp-docs ✗ ✗ via provider ✅ ~ ✗ ✗ ~ ✗ ✗ ✗ ~ ? ✅ (via @mcp-b)
contentstack-webmcp ✗ ✗ ✅ CDA ✅ ✗ ✗ ✅ ✗ Contentstack only ✗ ✅ ✅ ? ✅
auto-webmcp / keak ✅ forms only ✗ ✗ ✗ ✗ ✗ ✅ ✗ ✗ ✗ ~ ✅ ✅
Hand-rolled Hugo/Jekyll (Chaganti, pardel) ✗ ✗ titles/tags ✅ (.md) / ✗ ✗ ✗ ✅ 1 site ✗ ✗ ✗ ~ ✗ ✅
Vendor MCP servers (Sanity/Contentful/Strapi/WP adapter) ✗ ✗ ✅ ✅ ✗ ✗ ✅ ✗ own CMS own server admin creds ✅ not WebMCP
agentprose ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅ 6+ ✅ ✅ ✅ ✅ ✅
  1. Siloed per platform — Astro-only, WP-only, Contentstack-only; nothing spans static sites + WordPress + Ghost + headless CMS.
  2. No zero-config content discovery — nobody reads what every blog already publishes (JSON-LD BlogPosting, OpenGraph, RSS/Atom/JSON Feed, sitemap, llms.txt).
  3. Agents can’t actually read an article — astro-webmcp truncates every output at 1,500 chars; Jekyll example returns excerpts only; WP Bridge needs a second plugin for Markdown.
  4. No page-level tools — no “current article”, “table of contents”, “get section”, “cite/quote” — the thing an in-tab agent most needs.
  5. Index size is unsolved — Chaganti’s Hugo index hit 60 MB; hand-rolled sites ship one JSON blob; only astro-webmcp reuses Pagefind’s chunked index.
  6. Stale / wrong API — several plugins and posts target the deprecated navigator.modelContext (Agentgate, code-atlantic/webmcp-abilities, goSmarter post) and silently do nothing in current Chrome.
  7. Write-capable surface where read-only is the job — WP plugins mix cart/forms/admin; vendor MCP servers need admin tokens. Readers’ agents need read-only + untrustedContentHint.
  8. Token-leak footguns for headless CMS — no packaged direct/proxy split except an unpublished Contentstack experiment; no guard against shipping preview/management tokens.
  9. Zero observability — site owners can’t see which agents called which tools (Cloudflare/TollBit sell this for crawlers; nobody offers it for WebMCP calls).
  10. Untested, unmaintained hobby code — most WP WebMCP plugins have 0–10 installs; only Cloudflare and MCP-B have real engineering behind them; discoverability is an admitted limitation in Chrome’s own docs.

Legend: [P]=page/registry/API read this session; [S]=search snippet only (unverified).

Tool What it is Content tools for blogs? Notes
Cloudflare “WebMCP on Cloudflare” (dev preview, blog 2026-08-06) https://blog.cloudflare.com/webmcp/ [P] Dashboard toggle (Agent Readiness → WebMCP); edge HTMLRewriter injects <script type="module" src="/.webmcp/bridge.js" data-packs="c2pa,mcp-server-client" data-mcp-url="/mcp">; bridge registers tool packs via document.modelContext.registerTool; no-op when unsupported No content pack yet. Packs: Content Credentials (scan_images_c2pa, inspect_image_c2pa) and Site MCP Server (proxies your own /mcp tools/list → tools/call with the visitor’s session). Blog says future packs may “summarize a sitemap with Workers AI or query an AI Search index”. Biggest strategic threat/validation: proves the “one injected line” model. Cloudflare-only; requires an existing MCP server for anything site-specific. Also BrowserRun (remote browser) consumes WebMCP: https://developers.cloudflare.com/browser-run/features/webmcp/ [S]
MCP-B / WebMCP-org https://github.com/WebMCP-org/npm-packages [P] (104★, MIT, pushed 2026-10-14) Polyfill + transports + React hooks + local relay No blog tools — infrastructure @mcp-b/webmcp-polyfill ~702k dl/mo, @mcp-b/global ~439k/mo [P npm search API]. Bridges in-tab tools to desktop MCP clients (local relay). We should interoperate, not compete.

2b. WordPress (≈20 plugins mention WebMCP; all small) — wordpress.org plugin API queried [P]

Section titled “2b. WordPress (≈20 plugins mention WebMCP; all small) — wordpress.org plugin API queried [P]”
Plugin Installs Updated WebMCP tools / approach Strengths Weaknesses
Royal MCP https://wordpress.org/plugins/royal-mcp/ 10,000+ 2026-10-15 Server-side MCP (200+ tools, OAuth 2.0); “WebMCP-ready on the server side”: signed-in visitors’ browser agents via a WebMCP bridge (names Cloudflare’s) Breadth, OAuth, integrations (Woo, Elementor, Yoast…) Admin/write oriented; WebMCP only for signed-in users via third-party bridge; huge attack surface; not a reader-facing content layer
WebMCP Bridge https://wordpress.org/plugins/webmcp-bridge/ 600+ 2026-09-25 search_posts, get_post, get_menu, get_categories, get_site_info, submit_contact_form + 8 WooCommerce tools; manifest at /wp-json/webmcp-bridge/v1/manifest, OpenAPI, remote MCP at /wp-json/webmcp-bridge/v1/mcp; with paid-ish companion “Mescio for Agents”: retrieve (RAG), get_markdown_content, get_llms_txt Most complete reader tool set in WP; multi-surface (WebMCP + MCP + OpenAPI) Markdown/RAG need a 2nd plugin; WP-only; mixes write actions (cart)
Conversion Agent Discovery https://wordpress.org/plugins/conversion-agent-discovery/ 100+ 2026-06-17 read-only WebMCP + Accept: text/markdown, llms.txt, api-catalog, agent-skills, optional Origin-Trial header Honest, standards-heavy, read-only tiny adoption; WP-only
Agentgate for WebMCP https://wordpress.org/plugins/agentgate-for-webmcp/ 10+ 2026-04-02 3 read-only tools (search posts, get post, list categories); checks navigator.modelContext zero-config targets the deprecated navigator alias → silently dead on current Chrome; abandoned since April
BeeClear / Rootz / IBS Gateway / GoldT / Semantika / QuangNhatDev / Nurevo / RankReady(github eticastudio/rankready, 9★) 0–10 each 2026 Mostly “AI visibility/SEO” bundles: llms.txt, markdown, /.well-known files + some WebMCP tools lots of surfaces Fragmented, overlapping, near-zero installs; proprietary-ish “standards” (/.well-known/ai, agents.txt)
Form plugins (IBS for WPForms/Ninja, WebMCPFY CF7/Gravity, Form Annotator) 0 2026 Declarative form tools narrow not content
Official WordPress MCP Adapter https://github.com/WordPress/mcp-adapter [P] (1,849★, GPL-2.0, pushed 2026-10-16) + Abilities API (merged to core; repo archived https://github.com/WordPress/abilities-api 215★) — — Server-side MCP over the Abilities API; Automattic/wordpress-mcp (948★) archived/deprecated in favour of it Canonical, core-backed Server-side, authenticated, admin/editor use-case — not an in-page reader layer
code-atlantic/webmcp-abilities https://github.com/code-atlantic/webmcp-abilities [P] 14★ 2026-02-25 Bridges WP Abilities → navigator.modelContext smart idea (reuse Abilities) stale; deprecated navigator API

2c. Static site generators & JS frameworks

Section titled “2c. Static site generators & JS frameworks”
Tool Stats Approach Strengths Weaknesses
astro-webmcp https://www.npmjs.com/package/astro-webmcp [P] / github fabricioctelles/astro-webmcp (19★, MIT, v0.6.0 2026-09-19, ~788 dl/mo) Astro integration, “one line” in astro.config; tools search_content, list_sections, go_to, get_page_info; search backends manifest/Pagefind/Orama; security: exposedTo, maxOutputLength 1500, regex prompt-injection sanitizer; annotations Closest competitor; good security thinking; Pagefind reuse Astro only; no full-article/markdown tool (outputs capped at 1,500 chars); no TOC/sections, no related, no feeds/JSON-LD discovery; customTools as code strings
webmcp-docs https://www.npmjs.com/package/webmcp-docs [P] (MIT, v0.4.0 2026-09-01, ~58 dl/mo) Library: search_docs, get_doc, list_docs; you supply pages + provider(search,getDocument); depends on @mcp-b/global framework-agnostic docs-only; manual wiring; no auto-discovery
Hugo — hand-rolled: ravichaganti.com (5 tools, see §0) https://ravichaganti.com/blog/adding-webmcp-support-to-hugo-static-sites/ [P]; goSmarter.ai (Hugo/Netlify, uses navigator.modelContext) https://www.linkedin.com/pulse/getting-our-hugonetlify-site-agent-compatible-steph-locke-bnjoe [P] Per-site JS + Hugo JSON outputs proves demand No reusable Hugo module exists (GitHub search “webmcp hugo”: 0 repos [P])
Jekyll — hand-rolled: pardel.dev (Sep 28 2026) https://www.pardel.dev/2026/09/28/webmcp-this-blog-now-has-tools [P] /webmcp.json built by Liquid + one script: search-posts, list-tags, list-projects, open-page, list-talks; readOnlyHint clean, idiomatic bespoke; excerpts only (no full text); no Jekyll plugin exists (GitHub search: 0 repos [P])
Eleventy, Docusaurus, VitePress, Next.js, Gatsby, Nuxt Content, Hashnode, Medium, Substack, dev.to No dedicated WebMCP content integration found (npm + GitHub + web search) [P/S] — gap
Generic auto-generators: auto-webmcp (forms → tools, script tag, v0.7.0, MIT) https://www.npmjs.com/package/auto-webmcp [P]; @keak/webmcp-core (crawler → tool defs, MIT) [P]; webmcpable (app actions, MIT, v0.7.0 2026-10-15) [P]; @absolutejs/webmcp (policy/audit) [P npm] Forms/app-action oriented one-line install proven (auto-webmcp) Not content-aware: a blog has no forms worth calling; reading/searching articles is the job

2d. Headless CMS — vendors ship server-side MCP, nobody ships reader-side WebMCP

Section titled “2d. Headless CMS — vendors ship server-side MCP, nobody ships reader-side WebMCP”
  • Sanity, Contentful, Strapi, Storyblok, Hashnode, Ghost (community) all have MCP servers aimed at editors/developers (authenticated, write-capable, run in IDE/desktop) — see §3 for primary links. Strapi sidecar corsen-context-strapi (0★, MIT, 2026-09-03) https://github.com/CorsenAI/corsen-context-strapi [P] is the only CMS-side WebMCP attempt found.
  • Enonic: no MCP/WebMCP product found [S].
  • Gap: no package turns a headless CMS’s public delivery API into reader-facing in-page tools.

2e. Adjacent agent-readiness layers (not WebMCP but overlapping)

Section titled “2e. Adjacent agent-readiness layers (not WebMCP but overlapping)”
  • llms.txt plugins (WordPress “Website LLMs.txt”, Yoast llms.txt feature) and Cloudflare “Markdown for Agents” — static discovery/markdown, no tools. NLWeb (Microsoft) https://github.com/nlweb-ai/NLWeb [P] 6,255★ MIT — Python server: natural-language /ask over schema.org data + MCP; heavyweight (vector DB). Scoring sites: webmcp.ora.ai (referenced in webmcp-docs README) [P].

agentprose is a reader-side layer. It serves the agent in the visitor’s browser, from public content, read-only. The vendors’ MCP servers (Sanity, Contentful, Strapi, the WordPress MCP Adapter) serve the editor’s agent with admin credentials. The two are complementary. agentprose interoperates with MCP-B: it registers into the MCP-B polyfill and is visible to its relay.