How AI agents value hotels: the HotelVALORA MCP server

Why an MCP server instead of a chatbot

The Model Context Protocol (MCP) is the open standard that lets AI assistants call external tools. Instead of HotelVALORA shipping its own chatbot, any capable agent — claude.ai, ChatGPT in developer mode, Perplexity, Gemini, Claude Code, or an SDK-based agent you build — connects to the HotelVALORA MCP server and gains hotel valuation as a native capability.

The difference with asking a general-purpose model to "value a hotel" is the data and the discipline: the agent gets sealed, institution-grade analyses computed by HotelVALORA's valuation engine on real market data — not numbers the language model makes up.

The endpoint and authentication

MCP server (Streamable HTTP): https://www.hotelvalora.com/api/mcp

Always use the www host — the bare domain redirects, and MCP clients drop the Authorization header on redirects.

Two doors, one identity: OAuth 2.1 with discovery (RFC 9728 protected-resource metadata at /.well-known/oauth-protected-resource/api/mcp) and dynamic client registration — so clients like claude.ai connect with no keys to copy — or a hvk_ API key from your account used as a Bearer token for headless clients.

MCP access is included in Premium plans and above. Quota is measured in distinct assets analysed, not API calls: re-querying the same asset does not consume again.

The nine tools

Every tool is read-only or ephemeral compute: nothing an agent does mutates reports or account state.

• search_assets · get_asset — find assets and read their identity sheet (no quota consumed).
• get_report · get_compset — the full sealed analysis and the competitive set (consume one distinct asset; the same asset never re-consumes).
• get_market — market and submarket aggregates (census, ADR, occupancy, RevPAR) with no consumption.
• adjust_scenario · goal_seek — CAPEX what-if with its own receipt, and the entry price that achieves a target IRR.
• export_report — a link to the watermarked institutional PDF with a verifiable receipt.
• list_own_usage — your tier, consumption and limits at any moment.

Verifiable receipts: the trust mechanism

Each full analysis served to an agent is sealed with an immutable receipt: an HV-… code plus a sha256 hash of every figure served. Any reader can verify at hotelvalora.com/verify/<code> — no account needed — that the figures an AI quoted came from HotelVALORA and were not altered or hallucinated.

This closes the loop that plagues AI-generated analysis: an agent citing HotelVALORA can hand its reader a one-click proof. Sealed blocks carry validity dates and provenance as fields; agents are instructed to quote them as-is rather than recombine figures out of context.

Connecting, in practice

Example prompts once connected: «Find hotel Riu Plaza España and give me its full valuation with the verification code» · «What ADR and occupancy does the Madrid Centro market have?» · «Apply a €40,000-per-key CAPEX and tell me how value and IRR move».

• claude.ai (web and mobile): Settings → Connectors → add custom connector with https://www.hotelvalora.com/api/mcp — OAuth discovery and consent, no keys.
• ChatGPT (Plus/Pro): Settings → Apps → Advanced → Developer mode → add the MCP connector with the same URL.
• Perplexity (Pro/Max): Settings → Connectors → add custom connector. Gemini: gemini.google.com/apps → custom app with the server URL.
• Headless (Claude Code, Claude Desktop, MCP Inspector, your own SDK agent): pass the hvk_ API key as a Bearer token.

Scope and limits

• Asset types: existing hotels, hotel projects from an address (auto-fitted on official Spanish cadastre data), hostels, serviced apartments, and office/residential buildings valued in hotel use.
• Current institutional coverage: Madrid (Spain), expanding.
• Analyses are informational, not investment advice; exact figures for unverified readers follow the same banding rules as the public library.