Urgent.News

What's breaking now, across thousands of outlets.

AI

MCP for tour guides: building a remote MCP server with ~100 tools and OAuth

I work on Your Next Tours, an app tour guides use to broadcast live audio to their group from their phone. Guests scan a QR code and listen in the browser. Before a tour, guides prepare a lot of material in our panel: day-by-day programs, stops, info cards, guest lists, sometimes a small company website. Most of that material already exists somewhere else, usually as a PDF itinerary. So we built…

The author discusses the creation of a remote Multi-Modal Content Provider (MCP) server for tour guides. This server enables guides to connect their own AI models, such as Claude, ChatGPT, or Cursor, to their accounts. Instead of manually retyping material, guides can upload PDF itineraries, and the AI model processes them using tools. The source covers the setup process, challenges faced, and solutions implemented for the server.

The server is built with the endpoint https://api.yournext.tours/api/mcp/guide and uses Streamable HTTP for transport, which is stateless and runs on a PM2 cluster. Authentication is handled via OAuth 2.1 + PKCE for Claude.ai and ChatGPT, or a personal API key for clients sending only authorization headers. The authorization server is self-hosted Ory Hydra, which delegates login and consent to the existing account system.

The setup includes 100 tools, each with a specific scope. The LLM cost is covered by the user's own subscription, which is why the setup was beneficial for the creators. They also decided not to build their own OAuth server but to use Ory Hydra instead.

To handle stateless Streamable HTTP, the source server explicitly returns a 405 status code for GET and DELETE requests, as these methods do not support Server-Sent Events (SSE) streams. The 401 response code is crucial for the authorization process, as the authorization server's WWW-Authenticate header must include the scope.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

I'm building a companion island for my desktop AI assistant: approvals, a mascot, and a 14px wake strip

I'm building a desktop AI assistant, and it has a permission problem that no amount of prompt engineering fixes. Ankita asks for approval before it does anything irreversible — sends a message, runs a…

  • Reporter building companion island for desktop AI assistant
  • Island includes three views (Petit tab, Home, Home expanded) and 14-pixel wake strip
  • Keyless Composio sign-in and MCP approval tiers introduced alongside

A Local AI System Reliability Agent with Gemma 3 4B

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend What I Built I built a local AI system reliability agent that monitors a Windows computer, stores system metrics…

  • Local AI system reliability agent developed for Hacktoberfest Weekend Challenge
  • Gemma 3 4B local AI model used to analyze system trends and provide insights

Is your AI agent worth its tokens? We measured it with TigerGraph

Is your AI agent worth its tokens? We measured it with TigerGraph Tags: ai, rag, graph, python Everyone is bolting agents onto retrieval. Almost nobody asks what they cost.

  • Only 3% of questions benefit from AI agents according to TigerGraph study
  • OCCAM tool assesses AI agent value by routing questions through tiers
  • Tier 0 answers 98 questions with zero tokens, tier 1 costs around 830 tokens

More from Saturday 3 October →