THE GLASS LANTERN HOTEL
Fictional standalone OpenAI reception demo by Techie Stays.
Location: Kamala, Phuket, Thailand.
Website: https://glass-lantern-hotel.vercel.app/
Human voice/text reception: /reception
MCP endpoint: https://glass-lantern-hotel.vercel.app/api/mcp
Transport: Streamable HTTP, JSON-RPC POST. No SSE stream. MCP protocol 2025-11-25.

DISCOVERY AND CONNECTION
Find the hotel website from its name and location. Public pages have no booking credentials. With the guest's permission, privately submit the name and reference from their fictional confirmation; ota is optional. Never put booking details or tokens in search queries or URLs. This demonstration does not require OTA messaging.

MCP TOOLS
hotel_information: public fictional hotel facts.
connect_reservation: arguments name, reference, optional ota. Returns a 15-minute access_token. Keep it private.
message_reception: arguments access_token, message. Sends to the standalone OpenAI assistant.
read_conversation: argument access_token. Reads the shared demo reservation conversation.
check_in_information: argument access_token. Returns fictional arrival instructions and email-ready information. Does NOT complete check-in or send email.

HTTP ALTERNATIVE
POST /api/connect with JSON name and reference, optional ota. Keep returned access_token private.
GET /api/reception with Authorization: Bearer <access_token> returns a unique connection receipt.
POST /api/chat with the same Authorization and JSON message sends to the hotel assistant.
GET /api/chat with the same Authorization retrieves the conversation.
GET /api/check-in with the same Authorization retrieves a check-in guide and email draft.

All private tools refer to ONE fictional demo reservation and its shared conversation. Anyone with that fixture's valid access can read it. Use only fictional information. Chat is stored by OpenAI until deleted by the organiser. Browser voice input and playback are available where supported; speech may be processed remotely. No production identity verification.

PROOF
Discovery: reach this published site. Connection: retrieve the authenticated receipt. Conversation: send a new question or harmless surprise word, then retrieve the hotel's actual response and confirm it appears in website chat. Report each result separately. A scripted welcome is not evidence of AI conversation.

An MCP-capable app must be configured to use the endpoint. Merely reading this file does not grant tool capability. If an AI only reads websites, provide the human reception link and report discovery only. Do not claim actions without actual responses.

EMAIL
Prepare a draft for the guest to review. There is no email delivery integration. No payments, transfers, upgrades or check-in are confirmed by this demo.

WEBMCP IN THE BROWSER
Open /reception in a WebMCP-capable browser agent. The page registers hotel_information, connect_reservation, message_reception, read_conversation and check_in_information. Browser tools keep the session token within the page; pass only the guest-supplied name/reference to connect, then message text to chat. Tools update the visible reception. No automatic email delivery. The AI section shows whether this browser registered all five tools. This does not enable tools in ordinary read-only Gemini chat or fix search indexing.