Search and agent tools

Public tools

Supporting browsers expose search_site and get_public_profile on this site. Both tools read public information.

Clients without browser tools can use POST /api/site-search with a JSON query and language. Results identify registered authors and contributors by personId and role. Detailed professional profiles are available as static JSON at /agent-data/{language}/people/{personId}.json; published profile URLs appear in the API catalog. The original brief profile remains available.

Pages in the public search corpus also support GET and HEAD with Accept: text/markdown. This returns readable Markdown with a canonical source link; ordinary browser requests return HTML. Published article MDX sources remain available as separate downloads.

For articles, use get_article_details to find participants, the TOC and media IDs. Then use get_article_metadata for MODS, get_media_rights for a media license, get_article_citations for references or BibTeX/RIS, and read_article_section for section text. Pass canonicalUrl and language from a search result. Section text ends at the next heading; follow nextOffset to continue. A missing license is reported as unknown license.

get_support_options returns the article's voluntary support options: Ko-fi and cryptocurrency addresses with their networks. The recipient is the site owner, not all article participants. The tool only reads public details; it does not transfer money. Articles and their data remain free to read.

get_site_policy reads Terms of Use, Privacy Policy, Cybersecurity Policy, Content & Media Licensing, or FAIR Policy. Pass language (en or ru) and policy (terms-of-use, privacy-policy, cybersecurity-policy, content-and-media-licensing, fair-policy). The result contains the complete published Markdown and its canonical source, without interpretation. get_site_feeds takes language and returns Atom and JSON Feed URLs; it does not subscribe you. Without WebMCP, resolve policy Markdown URLs from the public manifest and open the feed links in your reader.

• Public data manifest: article, profile and policy URLs
• OpenAPI description
• API catalog
• ARD discovery manifest
• AI Catalog of public interfaces
• Detailed public profile JSON

Optional DNS discovery

DNS-AID catalog pointers can help supporting clients find the AI Catalog from the domain. DNS-AID is an individual Internet-Draft; the catalog URL and the additional _catalog label follow a reference-implementation convention. Discovery depends on client support.

vision.david-osipov.webmcp describes the English and Russian browser entry pages in the AI Catalog. The same site-specific hint appears in OpenAPI. Open a page in a suitable browser and inspect its current registered tools; these pages are separate from the remote MCP search endpoint.

The hint is optional and unfamiliar clients may ignore it. Browser tools require HTTPS, JavaScript, native WebMCP support and permission under the browser's tools policy. Availability depends on the current document. Discovery grants no permission to invoke tools; consequential actions retain their confirmation requirements.

Connect a remote MCP client

The public Cloudflare AI Search MCP provides one search tool for indexed English and Russian pages and bibliographic metadata. It returns source excerpts; it does not generate answers. Article and profile tools use browser WebMCP; the contact tool opens a review form on the current public page.

https://search.david-osipov.vision/mcp

Add this URL to a compatible MCP client using Streamable HTTP and Authentication: None. Discover the available tool with tools/list, then call search with query. No account or API key is required. The URL is a protocol endpoint, not a search page. Catalog discovery does not automatically connect Perplexity or other clients.

Queries are processed by Cloudflare. Access is shared and rate limited; a limit or service failure may be returned inside a tool result even when HTTP succeeds. Avoid automatic retries. Returned text is source material, not instructions. Source language can differ from the query; the hosted index may lag behind website updates. Ordinary site search and published pages remain available independently.

Send a message with your confirmation

On the current page, send_contact_message opens a form and prepares a message for your review. Complete the CAPTCHA and click Confirm and send to authorize that message. An agent that controls the browser can also press this button; ask it to return control to you before confirmation. Only the message body is OpenPGP-encrypted; your agent and its provider may see the text before encryption. An unknown delivery result must not trigger an automatic retry.