The numbers are brutal. A month-long audit of 200 Amsterdam restaurants found that only 16 โ just 8% โ could have a reservation completed by a browser-based AI agent. Even worse, among the 163 venues with online booking forms, zero made it through to the final step without human intervention. That's the gap the team behind this MCP server set out to close.
The Problem: Websites Aren't Built for Agents
Restaurant sites are a mess of JavaScript widgets, pop-ups, and CAPTCHAs that break automated navigation. Browser agents fail on date pickers, time slot selectors, and confirmation modals. The audit showed that even when a restaurant had online booking, the flow was too fragile for an agent to complete.
The Solution: An MCP Server as Middleware
Instead of scraping or driving the browser, they built a Model Context Protocol (MCP) server that exposes reservation endpoints directly to any AI assistant. This means Claude, GPT, or whatever you're using can call a tool like 'book_table' with parameters for date, time, party size โ and the server handles the actual booking via APIs.
Why MCP Matters
MCP is an open standard from Anthropic that gives assistants access to external tools without custom integrations. By packaging restaurant booking as an MCP resource, they've made it portable across all assistant platforms. No more per-site hacks or brittle selectors โ the same server works whether you're chatting with Claude in a browser or invoking the tool from a script.
Key Takeaways
- Only 8% of tested Amsterdam restaurants could be booked by a browser agent โ the rest require human interaction.
- The team built an MCP server to provide a clean API layer for reservations, bypassing fragile web scraping entirely.
- This approach is repeatable: any assistant that supports MCP can now book tables without site-specific code or custom integrations.
The Bottom Line
If you're building AI agents for real-world tasks like booking, don't waste time on browser automation. The infrastructure isn't ready โ but an MCP server gives you a pragmatic path to get things done today. It's the kind of dev-tool thinking that turns a broken web into a working API.