← Home

For agents

Project

Public, read-only capabilities and connection details for AI agents using Mayphus.

Search the full published text, read a canonical document, and follow its sources. The website, plain text, HTTP API, and MCP gateway share the same public content.

Read directly

No connection is needed for these ordinary HTTP requests. Search returns stable IDs and excerpts; reading returns full text, a citation URL, and metadata.

GET /api/search?q=bootstrapping
GET /api/document?url=/a-tiny-freebsd-jail-doc-fix/

Document responses also include derived assets and relations when present. Selected visual records have a model: boot stages, tile studies, source-table rows, audio samples or links to the exact public dataset used by the browser. The same objects appear in the corpus and MCP fetch; they add no write operation. The original narrative and evidence qualifications remain in text.

Read the complete public document corpus or a plain-text example.

One public, read-only gateway

Connect explicitly to the public MCP server. It exposes published material and bounded tools; it does not grant access to private site areas or change site content.

https://mayphus.org/mcp

Nothing here silently installs or connects an agent. A person or developer chooses to add this endpoint. The public HTTP interfaces also work directly.

Browser site tools (WebMCP)

In a browser supporting the proposed WebMCP API, the ordinary reading pages register four read-only tools in the top-level document:

Tool Behavior
search_site Search /agent-index.json metadata (title, summary, tags) with an optional language filter; return up to eight matches. Quiet archive entries stay excluded.
read_page Read one published route as plain text using the same non-browser Accept representation the Worker already serves.
current_page Return the open page title, canonical URL, summary, and plain-text body.
get_profile Return the public profile and résumé links from /api/profile.

All four reuse same-origin public endpoints only. They reject unexpected fields, invalid inputs, and external URLs. They expose only the published corpus; they do not access private work, unpublished records, or private datasets.

Tools register when document.modelContext.registerTool is available (with a trivial navigator.modelContext fallback where present). Registration is scoped to the page lifetime and renewed when the page returns from the back/forward cache. Unsupported browsers keep the ordinary links and reading interface. No polyfill, account, new connection, or credentials are needed. Browser support and agent discovery vary; this is progressive enhancement against the current draft, not a universal conformance claim.

References: ChatGPT site tools and the WebMCP draft.

Capabilities

These six public capabilities cover discovery, reading and bounded map generation. The table lists the HTTP endpoints and corresponding MCP tools; the capability links lead to the human reading views. On a narrow screen, scroll the table horizontally to read all columns.

Capability HTTP API MCP tools
Discover public work: published capabilities, projects and pages. /api/capabilities, /api/projects, /api/pages discover_capabilities, discover_projects
Fetch a public record: full text and source metadata, by stable ID or canonical URL. /api/document fetch, get_project
Follow public updates: recent articles and notes through the RSS feed, without inferring private activity. /api/updates, /api/notes get_public_updates
Generate a city map: a bounded city-and-topic request returning an OpenStreetMap-based result without changing server state. /api/map generate_city_map
Read the public profile: deliberately published profile, names, links and selected work. /api/profile, /api/identity search, fetch

For the exact map parameters, an HTTP request and MCP arguments, see the bounded map example. That page also records the prototype’s upstream availability limits.

Connect

Keep tool approval enabled when trying a new remote MCP server, even when its published tools are read-only.

ChatGPT

Create a custom MCP app or connector and use this server URL:

https://mayphus.org/mcp

Codex

Add this remote server to the Codex MCP configuration:

[mcp_servers.mayphus]
url = "https://mayphus.org/mcp"

Responses API

Place this object in the request’s tools array:

{
  "type": "mcp",
  "server_label": "mayphus",
  "server_url": "https://mayphus.org/mcp",
  "require_approval": "always"
}

Public discovery

These indexes are ordinary public files. They make the site easier to inspect, cite, and reuse, but they are not an automatic authorization mechanism.

Copy as text