# gabrimatic.info / agent directory

Agent Discovery

Every URL on this page is a live, machine-readable endpoint. Open any of them in a terminal, an MCP client, or an A2A-aware agent. There is no auth and no rate limit beyond GitHub's own anonymous quota for the proxied repository endpoint.

100 / 100

Level 5 · Agent-Native

Verified by isitagentready.com. Re-scanned after every deploy; all 12 scored checks pass on every release.

Real-time protocols

Model Context Protocol · Streamable HTTP, JSON-RPC 2.0 POST https://www.gabrimatic.info/api/mcp: Stateless MCP server. Implements initialize, ping, tools/list, tools/call, resources/list, resources/read, and prompts/list. Tools: get_profile, get_journey, get_repos, get_page_markdown. Discovery card at /.well-known/mcp/server-card.json.
Google Agent-to-Agent (A2A) · JSON-RPC 2.0 POST https://www.gabrimatic.info/api/a2a: Stateless A2A endpoint. Implements message/send, tasks/get, and tasks/cancel. Returns a synchronous completed Task whose artifact lists every authoritative knowledge source. Agent Card at /.well-known/agent-card.json.
WebMCP · in-page tool registration navigator.modelContext on https://www.gabrimatic.info/: When the homepage loads in an MCP-aware browser, /webmcp.js registers navigate, get_page_markdown, and list_repos via navigator.modelContext.registerTool. Spec: webmachinelearning.github.io/webmcp.

JSON API

Resource	URL	Cache
Profile	`/api/profile`	1 hour
Professional journey	`/api/journey`	1 hour
GitHub repositories (live)	`/api/repos`	edge 5 min, SWR 5 min

/api/repos accepts ?per_page= (1 - 100, default 100) and ?sort= (updated, created, pushed, full_name; default updated). All endpoints set Access-Control-Allow-Origin: *.

OAuth 2.1 authorization server

A real OAuth 2.1 authorization server is exposed for agents that need bearer tokens (for example, MCP clients that require authentication before tool calls). All read endpoints on this site also allow anonymous access, so OAuth is optional. Dynamic Client Registration is enabled: any agent can self-register.

Endpoint	URL	Spec
Authorization server metadata	`/.well-known/oauth-authorization-server`	RFC 8414
OpenID Connect discovery	`/.well-known/openid-configuration`	OIDC Core
Protected resource metadata	`/.well-known/oauth-protected-resource`	RFC 9728
JWKS (EdDSA public key)	`/.well-known/jwks.json`	RFC 7517
Web Bot Auth directory	`/.well-known/http-message-signatures-directory`	Web Bot Auth + RFC 9421
Dynamic client registration	`POST /api/oauth/register`	RFC 7591
Authorization endpoint	`GET /api/oauth/authorize`	RFC 6749 + PKCE S256 (OAuth 2.1)
Token endpoint	`POST /api/oauth/token`	RFC 6749 + OAuth 2.1

Access tokens are EdDSA-signed JWTs (typ: at+jwt), iss = https://www.gabrimatic.info, aud = https://www.gabrimatic.info/api/mcp, 1 hour TTL. Supported scopes: mcp:read, mcp:tools.

Markdown for Agents

Every page route returns Markdown when the request carries Accept: text/markdown. Direct *.md URLs are equivalent. Vary: Accept is set on every negotiated route.

curl -s -H 'Accept: text/markdown' https://www.gabrimatic.info/journey

Route	Markdown alternate
`/`	`/index.md`
`/journey`	`/journey.md`
`/interests`	`/interests.md`
`/opensource`	`/opensource.md`
`/definition`	`/definition.md`

Discovery

Agent Skills index · v0.2.0 /.well-known/agent-skills/index.json: Lists each SKILL.md with a SHA-256 digest so agents can cache and verify.
API Catalog · RFC 9727 /.well-known/api-catalog: Linkset advertising every API, discovery card, and Markdown alternate exposed by the site.
Sitemap /sitemap.xml: Standard XML sitemap. Also advertised via RFC 8288 Link headers on /.
RFC 8288 Link headers HEAD /: The home route returns Link headers advertising the sitemap, the Agent Skills index, the A2A agent card, the MCP server card, OAuth AS metadata, the OAuth protected resource, the API catalog (RFC 9727), the HTTP message signatures directory, and the Markdown alternate (9 values total, all rel=describedby except sitemap and alternate).
Content Signals /robots.txt: Content-Signal: search=yes, ai-input=yes, ai-train=no: opted in to retrieval and answering a user's question right now, opted out of training corpora. Training-only crawlers are explicitly blocked by user-agent.