Conversion Agent Discovery
| 开发者 | agenciaconversion |
|---|---|
| 更新时间 | 2026年6月18日 06:00 |
| PHP版本: | 7.4 及以上 |
| WordPress版本: | 7.0 |
| 版权: | GPLv2 or later |
| 版权网址: | 版权信息 |
详情介绍:
Conversion Agent Discovery helps WordPress sites expose public, read-only discovery surfaces for AI agents without pretending to support capabilities that are not implemented.
It is designed to improve measurable agent accessibility for public, read-only discovery checks without calling external scanners automatically.
Features include:
- Markdown negotiation for public content with
Accept: text/markdown. - Content Signals in
robots.txt. - Generated
llms.txtand/.well-known/llms.txt. - Editable AI Instructions page for official brand facts and LLM guidance, starting empty with English templates by default.
- Runtime Front-end Placement for exact, design-safe AI Instructions link placement across matching public header, footer, or page content components.
[conversion_ai_instructions_link]shortcode for manual placement.- Generated
/.well-known/api-catalog. - Generated
/.well-known/agent-skills/index.jsonand virtualSKILL.mdfiles. - Local Agent Skills discovery schema generated by the site.
- Read-only WebMCP tool registration with runtime diagnostics for compatible browsers.
- Production-oriented Agentic Browsing readiness guidance for PageSpeed and Lighthouse checks.
- Optional origin isolation headers and optional Origin-Trial header for browser WebMCP runtime support.
- Opt-in declarative WebMCP form annotations for exact rendered form opening tags.
- Public read-only REST endpoints for site search, recent content, single content reads, site context, and contact handoff.
- Agentic Browsing diagnostics that separate plugin-controlled discovery surfaces from theme accessibility and layout stability checks.
- Admin settings page with a global public switch.
- Chris Long's LinkedIn post: https://www.linkedin.com/posts/chris-long-marketing_holy-moly-seos-ai-instruction-pages-influence-share-7466096236799053824-_rpb/
- Nectiv case study: https://nectivdigital.com/blog/aeo-experiment-chatgpt-cited-an-ai-instructions-page-in-48-hours
- Nectiv AI Instructions example: https://nectivdigital.com/ai-instructions
- Seer Interactive AI Information example: https://www.seerinteractive.com/ai-information
- Omniscient Digital AI Info example: https://beomniscient.com/ai-info/
- llms.txt proposal by Jeremy Howard: https://llmstxt.org/
安装:
- Upload the plugin ZIP in Plugins > Add New > Upload Plugin.
- Activate Conversion Agent Discovery.
- Open Settings > Conversion Agent Discovery.
- Review metadata and enabled modules.
- Purge site cache after activation.
常见问题:
Does this change the HTML site?
No. Normal browser and crawler requests continue to receive HTML. Markdown is returned only when the request explicitly sends Accept: text/markdown.
Does this create an MCP server?
No. Conversion Agent Discovery publishes only real read-only resources. It does not publish an MCP Server Card unless a real MCP server exists.
What does WebMCP support mean in this plugin?
WebMCP support means browser-side read-only tool registration on public pages when a compatible browser runtime exposes document.modelContext or navigator.modelContext. It is not a remote MCP server, transport, auth flow, or server capability advertisement.
Why is WebMCP origin isolation opt-in?
Browser WebMCP requires origin isolation headers. The plugin can send them, but this is opt-in because COOP and COEP can affect third-party embeds, OAuth popups, payment widgets, ads, and analytics.
Some Chrome/Lighthouse builds still report WebMCP audits as not applicable when the browser runtime does not expose document.modelContext or navigator.modelContext; in that case the local gate treats not-applicable WebMCP audits as non-failing while still failing score 0 or audit errors. For strict WebMCP validation, use Chrome 150 or newer with the required Chrome runtime support and run the assertion with CAD_REQUIRE_WEBMCP=1.
Does this annotate every form automatically?
No. Declarative WebMCP form annotations are disabled by default and require an administrator to opt in with an exact rendered form opening tag, tool name, and description. This read-only release does not add automatic form submission attributes.
Does this guarantee a perfect PageSpeed or Lighthouse score?
No. The plugin prepares the discovery surfaces it controls. Lighthouse Agentic Browsing also checks theme and site behavior such as accessibility tree quality and layout shifts.
Why does PageSpeed still report an Agentic Browsing accessibility issue?
The plugin cannot automatically repair markup produced by the active theme, menu, media library, or page builder. For example, a linked logo image with empty alt text can fail "Links must have discernible text". Fix that in the theme or page builder by adding meaningful image alt text, an aria-label on the link, or visually hidden link text.
Does the AI Instructions feature change the front-end automatically?
Only when Front-end Placement is enabled and the configured snippet is found in rendered frontend HTML. The plugin does not edit saved page content, theme files, or templates.
Does this guarantee ChatGPT, Claude, Gemini, Perplexity, or Google AI citations?
No. Results may vary by model, search index, prompt, location, account context, retrieval behavior, and crawl timing.
更新日志:
0.4.0
- Add production-oriented Agentic Browsing readiness guidance for PageSpeed and Lighthouse checks.
- Keep Agent Discovery as a protected brand name in bundled translations.
- Add optional Origin-Trial header support for Chrome WebMCP origin trial testing.
- Add canonical
wp_WebMCP tool aliases while preserving the legacy tool names for compatibility. - Add read-only
wp_list_termsandwp_list_sitemap_urlsWebMCP tools backed by public same-origin REST endpoints. - Add rate limiting to public read-tool REST routes without storing raw IP addresses.
- Exclude private, unpublished, password-protected, and cross-origin content from public read-tool REST responses.
- Add local activation guidance for Lighthouse Agentic Browsing, Chrome 150+ strict WebMCP validation, compatible Chrome binaries, and
CHROME_PATH. - Add strict
CAD_REQUIRE_WEBMCP=1assertion mode for local Lighthouse reports. - Add a demo checklist covering
llms.txt, API Catalog, Agent Skills, WebMCP, opt-in origin isolation, and declarative form annotations. - Add Agentic Browsing diagnostics for the experimental Lighthouse/PageSpeed agentic-browsing category.
- Prefer
document.modelContextfor WebMCP while keeping the legacynavigator.modelContextfallback. - Add optional origin isolation headers for browsers that require them before exposing WebMCP APIs.
- Constrain WebMCP input schemas with explicit
additionalProperties: false. - Add opt-in declarative WebMCP form annotations for exact rendered form opening tags.
- Add local Lighthouse Agentic Browsing scripts and a Docker fixture setup helper.
- Remove public automatic form submission output from declarative form annotations.
- Clarify that Agentic Browsing checks do not guarantee Lighthouse/PageSpeed scores, LLM citations, search rankings, AI Overview visibility, or model behavior.
- Reorganize the admin screen into Overview, AI Instructions, Discovery Surfaces, and Diagnostics tabs.
- Start AI Instructions empty and unpublished by default, with a generic English recommended structure and copyable English LLM prompt.
- Replace Footer Replacement with visual Front-end Placement for header, footer, or page content snippets while keeping legacy settings compatible.
- Add same-origin visual component selection, rich HTML editing, AI Instructions link insertion, and one-click placement publishing.
- Force Front-end Placement to use the homepage as the visual editing source, apply matching replacements across public pages, and normalize the legacy
{link_test}placeholder typo. - Stop advertising AI Instructions in llms.txt, Link headers, API Catalog, and Agent Skills until content is published.
- Replace the direct
wp_footerappend behavior with configurable runtime Footer Replacement. - Add source page import preview, exact snippet selection, custom replacement template, and configurable link text.
- Keep the default visual output restrained by linking only the
AI Instructionstext.
- Add editable AI Instructions page with accented Portuguese default content for conversion.com.br.
- Add optional footer link and
[conversion_ai_instructions_link]shortcode. - Add AI Instructions discovery links to
llms.txt, Link headers, API Catalog, diagnostics, and Agent Skills.
- Add bundled translations for 10 WordPress locales: Brazilian Portuguese, European Portuguese, Spanish, French, German, Italian, Dutch, Russian, Japanese, and Simplified Chinese.
- Add a generated POT catalog and load bundled translation files from
/languages. - Move WebMCP tool descriptions into translatable PHP configuration while keeping tool names and endpoints stable.
- Serve the Agent Skills discovery schema locally from the WordPress site.
- Remove non-essential external measurement links from the WordPress.org package.
- Clarify the External Services documentation for manual links only.
- Rename the plugin package to Conversion Agent Discovery with slug and text domain
conversion-agent-discovery. - Move admin CSS, admin notices JavaScript, and WebMCP JavaScript to local enqueued assets.
- Replace legacy WP Agentic identifiers with Conversion Agent Discovery prefixes.
- Use
rest_url()for REST API discovery URLs. - Add WordPress.org external services documentation.
- Prepare WordPress.org submission metadata, privacy notes, development source link, and package exclusions.
- Schedule rewrite flushes after plugin upgrades so third-party sitemap rules remain registered.
- Keep third-party WordPress admin notices out of the custom dashboard hero.
- Document measurement support for Is It Agent Ready and Agent Crawl by Conversion.
- Rename the public plugin name and distribution slug to Conversion Agent Discovery for WordPress.org directory compatibility.
- Add Conversion branding assets to the admin header and footer.
- Update plugin credit links to conversion.ag.
- Keep existing settings keys for smoother upgrades from earlier builds.
- Redesign the admin settings page as a richer dashboard with module explanations, scanner impact, diagnostics, and version footer.
- Add WPGraphQL status detection and installation guidance when WPGraphQL is not detected.
- Document why WebMCP is different from an MCP Server Card and keep fake OAuth, MCP Server Card, A2A, and commerce metadata unpublished.
- Add read-only WebMCP tools.
- Add public read-only Conversion Agent Discovery REST endpoints.
- Update Agent Skills discovery to v0.2 with virtual SKILL.md artifacts and SHA-256 digests.
- Add Markdown frontmatter and Content-Signal response headers.
- Add explicit Link headers for agent resources.
- Initial release.