Live music touring requires careful route planning. Independent musicians who book their own paid shows must connect venues between distant towns without wasting time or gas. When planning a run across multiple states, an artist cannot simply search a single city. They need to query stages along a highway drive, check room capacities, verify live music nights, and reach the person who books each room.

The Model Context Protocol (MCP) gives large language models a clean way to call external tools and retrieve structured records. Early MCP servers used local standard input and output (stdio) pipes. Modern cloud deployments use stateless Streamable HTTP transport. This guide walks through connecting a stateless Streamable HTTP MCP server with OAuth 2.1 to Claude Desktop, ChatGPT, and Google Gemini to research highway tour routes and small club stages.

Understanding Stateless Streamable HTTP in MCP

Traditional MCP setups run as local background processes communicating over stdio. While stdio works well on a single developer laptop, team workflows and cloud models need a web-accessible endpoint.

The MCP specification defines Streamable HTTP as an open web transport. Instead of keeping a stateful TCP connection or WebSocket open, each request travels over standard HTTP POST requests with JSON-RPC payloads. The server responds with application/json or streaming events (text/event-stream).

Stateless Streamable HTTP has three major operational benefits:

  1. Zero server memory leaks: Each HTTP request stands alone. The worker server does not store in-memory session tables.
  2. Edge deployment: The endpoint runs on global edge networks like Cloudflare Workers. Cold starts stay near zero milliseconds.
  3. Multi-client support: Desktop tools, browser interfaces, and mobile LLM apps can query the exact same MCP endpoint.

You can inspect a live implementation at the stateless Streamable HTTP MCP endpoint at No Stairs Please, which exposes venue and route tools across all fifty US states.

OAuth 2.1 Security for Stateless Tools

When exposing an MCP server to web-based AI clients, you must protect user data without adding server-side session state. OAuth 2.1 simplifies this architecture:

  1. Authorization Code Flow with PKCE: The client starts authentication by opening the authorization URL with a code challenge.
  2. Stateless Access Tokens: After user consent, the server issues a signed HMAC-SHA256 bearer token. The token payload encodes the user identifier, access tier, and expiration timestamp.
  3. Header Validation: When an LLM client calls a tool, it sends the token in the standard HTTP Authorization header: 
POST /mcp HTTP/1.1 Host: api.example.com Content-Type: application/json Authorization: Bearer <signed-token> { "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "search_venues", "arguments": { "stops": ["Seattle, WA", "Spokane, WA", "Missoula, MT", "Bozeman, MT"], "within_miles": 25, "max_capacity": 300 } } }

If an unauthenticated request hits a restricted action, the server returns an HTTP 401 status with a standard challenge header:

WWW-Authenticate: Bearer error="invalid_token", error_description="Token expired or missing"

Tool Definitions for Highway Corridor Routing

Tour routing requires geographic tools that understand road networks rather than simple radial distance circles. The venue MCP server provides the following tool signatures:

1. search_venues

Queries small stages by keyword, state, regional drive-time area, or highway road corridor. It accepts an ordered list of stops:

{ "name": "search_venues", "description": "Search live music venues across all 50 US states along a driving route corridor or town.", "parameters": { "type": "object", "properties": { "stops": { "type": "array", "items": { "type": "string" }, "description": "Ordered list of 2 to 15 towns along a highway driving route." }, "within_miles": { "type": "number", "description": "Corridor buffer width in miles (default 25)." }, "venue_type": { "type": "string", "description": "Category such as 'Music venue', 'Bar & pub', or 'Brewery & taproom'." }, "max_capacity": { "type": "number", "description": "Maximum room capacity limit." }, "music_days": { "type": "array", "items": { "type": "string" }, "description": "Filter by usual live music days of the week." } } } }

2. get_venue and reveal_venue

Retrieves stage dimensions, soundboard specs, past performance history, and direct talent buyer contact details for a specific stage.

3. find_by_performer

Looks up which stages similar touring bands have played along the same geographic corridor.

Connecting Claude Desktop

Claude Desktop connects to remote Streamable HTTP MCP servers using the local mcp-remote bridge or direct HTTP server blocks in your configuration file.

Open your Claude configuration file:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the remote server configuration:

{ "mcpServers": { "venue-router": { "command": "npx", "args": [ "-y", "mcp-remote", "https://api.example.com/mcp" ] } } }

Restart Claude Desktop. The hammer icon appears in the bottom right corner of the prompt box. You can now prompt Claude:

"Plan a 4-city tour starting in Seattle, WA and driving through Spokane, WA, Missoula, MT, and Bozeman, MT. Find rooms under 300 capacity that host live music on Thursday, Friday, and Saturday nights."

Claude invokes search_venues with the ordered stop list, maps the road corridor, and outputs a daily routing schedule with matching venues.

Connecting ChatGPT and Custom GPTs

OpenAI supports external tools through Custom GPT Actions and the OpenAI Apps SDK.

Setting Up a Custom GPT Action

  1. Go to the ChatGPT GPT Builder and open the Configure tab.
  2. Scroll down to Actions and click Create new action.
  3. Under Authentication, select OAuth 2.0.
  4. Set the Client ID, Authorization URL (https://auth.example.com/oauth/authorize), and Token URL.
  5. In the Schema box, paste the OpenAPI specification matching your MCP tools: 
openapi: 3.1.0 info: title: Live Music Venue MCP Routing API version: 1.0.0 servers: - url: https://api.example.com paths: /mcp: post: operationId: searchVenuesRoute summary: Search venues along highway stops requestBody: required: true content: application/json: schema: type: object properties: stops: type: array items: type: string within_miles: type: number responses: '200': description: List of venues matching the corridor

When a user in ChatGPT asks for tour stops, ChatGPT handles OAuth consent once, stores the bearer token securely, and queries the endpoint.

Connecting Google Gemini via Function Calling

Google Gemini models connect to MCP endpoints through the Gemini Python SDK using custom tool wrappers.

Here is a complete Python script showing how to wrap the Streamable HTTP MCP server into Gemini function declarations:

import os import requests import google.generativeai as genai genai.configure(api_key=os.environ["GEMINI_API_KEY"]) def query_mcp_tool(tool_name: str, arguments: dict) -> dict: url = "https://api.example.com/mcp" headers = { "Content-Type": "application/json", "Authorization": f"Bearer {os.environ.get('MCP_BEARER_TOKEN', '')}" } payload = { "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": tool_name, "arguments": arguments } } response = requests.post(url, json=payload, headers=headers, timeout=15) response.raise_for_status() return response.json().get("result", {}) def search_venues_along_tour(stops: list[str], within_miles: int = 25) -> dict: """Search music venues along a multi-stop highway corridor.""" return query_mcp_tool("search_venues", { "stops": stops, "within_miles": within_miles }) model = genai.GenerativeModel( model_name="gemini-1.5-pro", tools=[search_venues_along_tour] ) chat = model.start_chat(enable_automatic_function_calling=True) response = chat.send_message( "Find venues along Interstate 90 from Seattle to Bozeman with stops in Spokane and Missoula." ) print(response.text)

Gemini examines the tool schema, detects the multi-stop highway intent, calls search_venues_along_tour, and formats the final tour itinerary.

Handling Edge Cases in Highway Tour Routing

When building autonomous tour-booking agents, account for these real-world data constraints:

  • Drive-Time Corridor Width: Default to twenty-five miles. Rural mountain passes often have stages set five to ten miles off the main highway, while dense metropolitan areas require narrow corridors to avoid pulling irrelevant cross-city traffic.
  • Booking Lead Times: Small rooms book four to twelve weeks in advance. Direct your agent to check booking windows before suggesting routing dates.
  • Contact Separation: Always separate talent buyer contacts from general venue rental or wedding inquiry emails. Pitching a private event coordinator for a Friday night club slot will bounce or delay your booking email.

Conclusion

Stateless Streamable HTTP provides a robust, fast foundation for Model Context Protocol servers. By pairing OAuth 2.1 authentication with clean JSON-RPC tool definitions, you can expose rich geographic and business data to Claude Desktop, ChatGPT, and Google Gemini simultaneously. Musicians save days of manual searching, and agents deliver structured tour itineraries ready for direct booking outreach.

AI agents wrote this post and checked its facts against our own listings and sources.