A2A Travel MCP Servers (Remote)

Two standalone Model Context Protocol servers, each exposing one real-data tool over Streamable HTTP — deployable as independent remote services and reachable by any MCP client over a plain URL, instead of being spawned as a local subprocess.

Tools

Related MCP server: flights-mcp

Why Streamable HTTP

Both servers use the same tool logic you'd find in a local (stdio) MCP server — the only difference is the transport. Instead of an agent spawning these as a local subprocess, each one runs as a persistent web service with its own URL, so any MCP client (local or remote) can connect to it directly.

Running Locally

python3 -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt

FLIGHTAPI_KEY=... python flight_price_server.py    # listens on http://localhost:8000/mcp
RAPIDAPI_KEY=... python hotel_price_server.py       # listens on http://localhost:8000/mcp

Run one at a time locally (both default to port 8000), or set PORT to run them side by side:

PORT=8001 RAPIDAPI_KEY=... python hotel_price_server.py

Deploying to Render

Each server is deployed as its own Web Service, both pointing at this same repository:

1. Push this repo to GitHub.

2. On Render, click New + → Web Service, connect this repo.

3. For the flight service:

   • Build Command: pip install -r requirements.txt

   • Start Command: python flight_price_server.py

   • Add environment variable FLIGHTAPI_KEY with your key from api.flightapi.io/register.

4. For the hotel service, repeat as a separate Web Service:

   • Build Command: pip install -r requirements.txt

   • Start Command: python hotel_price_server.py

   • Add environment variable RAPIDAPI_KEY with your key from rapidapi.com (subscribed to the free tier of Booking COM).

Render assigns a public HTTPS URL to each service and injects PORT automatically — both servers already read it (os.environ.get("PORT", 8000)).

Connecting an Agent

Once deployed, point any MCP client at the service's /mcp endpoint. With the OpenAI Agents SDK:

from agents.mcp import MCPServerStreamableHttp

flight_mcp = MCPServerStreamableHttp(
    name="flight-price-server",
    params={"url": "https://<your-flight-service>.onrender.com/mcp"},
)

No local process is spawned — the agent talks to the tool entirely over HTTP.