Agent integration

Stsezhka exposes a compact public layer for assistants, search systems, and browser agents. It is designed to make product guidance easy to discover without turning private rider data into an agent-accessible API.

One boundary, everywhere

The public agent layer contains Stsezhka-authored documentation only. It cannot read an account, saved routes, activities, garage records, private shares, precise rider locations, credentials, or Strava data.

Start with the smallest useful document

Need	Canonical resource	Format
Curated site index	/llms.txt	Plain text / Markdown
Single-file product context	/llms-full.txt	Plain text / Markdown
Agent access policy	/agents.md	Markdown
API discovery	/.well-known/api-catalog	RFC 9727 linkset JSON
Public contract	/openapi.json	OpenAPI 3.1 JSON
MCP discovery	/.well-known/mcp/server-card.json	MCP Server Card JSON
Agent Skills	/.well-known/agent-skills/index.json	Agent Skills Discovery v0.2

The homepage also advertises the catalog, OpenAPI description, service documentation, llms.txt, and MCP Server Card through RFC 8288 Link response headers.

Markdown negotiation

Agents can request the homepage as Markdown:

curl https://stsezhka.com/ \
  -H 'Accept: text/markdown'

The Markdown response includes:

• Content-Type: text/markdown; charset=utf-8;
• Vary: Accept, Origin;
• X-Markdown-Tokens with a conservative token estimate;
• Content-Signal: ai-train=no, search=yes, ai-input=yes.

HTML remains the default for browsers.

Read-only MCP

The endpoint at https://stsezhka.com/mcp implements stateless Streamable HTTP and JSON-RPC 2.0. It supports the MCP 2025-06-18 protocol version and does not create a user session.

Tools

Tool	Purpose	Side effects
get_stsezhka_guide	Return the public product overview, canonical links, and privacy boundary	None
get_stsezhka_feature_guide	Explain one public planning or documentation workflow	None

Resources

• stsezhka://public/guide
• stsezhka://public/discovery

All tool responses are static, product-authored content. They do not read the current page, DOM, cookies, local storage, Firebase state, backend account records, or third-party services.

Minimal handshake

curl https://stsezhka.com/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  --data '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2025-06-18",
      "capabilities": {},
      "clientInfo": {"name": "example-client", "version": "1.0.0"}
    }
  }'

Browser-originated requests are accepted only from the canonical Stsezhka origin. Non-browser MCP clients may omit Origin.

WebMCP

When a browser implements WebMCP, the page registers the same two read-only tools on load. Stsezhka supports the current document.modelContext.registerTool() surface and the earlier navigator.modelContext experimental surface when present.

Both tools are annotated as read-only and return only trusted, product-authored strings.

Agent Skill

The Agent Skills index follows discovery schema 0.2.0. The stsezhka-public-guide artifact is served directly as SKILL.md and includes a SHA-256 digest of its raw bytes so clients can verify integrity and cache it safely.

The skill tells agents to:

1. begin with llms.txt;
2. use only the public MCP tools;
3. send riders to the visual planner for route creation;
4. keep authenticated and third-party fitness data outside agent flows;
5. treat route results as planning estimates, not safety guarantees.

Authentication is intentionally absent

Stsezhka does not publish OAuth/OIDC discovery, Protected Resource Metadata, or auth.md agent registration. Those protocols would imply that agents may access authenticated user APIs, which the product does not offer.

Do not infer access from a readiness score

A scanner score is a discovery heuristic, not permission to access private APIs. An absent auth manifest is an intentional security and compliance boundary until a separately reviewed, agent-safe authorization product exists.

Content and crawler policy

Public crawlable content declares:

• AI training: no;
• search use: yes;
• agent/AI input: yes.

robots.txt excludes API, routing, account, library, garage, activity, trip, route, and private share paths. Public search and agent use do not override those exclusions.

Rider safety

Routes, elevation, surfaces, timing, and directions are estimates. Assistants should remind riders to verify trail access, closures, weather, traffic, construction, and local rules before using a plan outdoors.