mcp-moz

A small MCP playground for demoing at MozFest 2026: an MCP server with a UI extension (MCP Apps), and a chat app that acts as an MCP host — it talks to a chat model, lets the model call the server's tools, and renders any ui:// resource (a Chart.js chart) inline in the conversation.

Components

Path	What it is
server.py	MCP server (streamable-http, port 8000) with tools over Barcelona 2025 vehicle data. Includes an MCP Apps extension binding the barchart tool to the ui://charts/bar.html Chart.js resource.
app.py	Starlette chat app (port 8080). Agent loop: Model → MCP tool calls → final reply. UI-bound tool results are returned as artifacts for inline rendering.
index.html	Chat front end (plain HTML/JS, no CSS). Loads ui:// resources in sandboxed iframes and delivers tool results via notifications/call_tool_result.
vehicles-2025.csv	Sample data (Barcelona Open Data).

Setup

Requires Python 3.14+. Uses uv:

uv sync

Create a .env file (used by the chat app; add .env to .gitignore):

API_KEY=<your API key>
AI_MODEL=<your AI MODEL>
AI_BASE_URL=<your AI Base URL>
MCP_SERVER_URL=http://localhost:8000/mcp

Run

# 1. MCP server (terminal 1)
uv run mcp run --transport streamable-http server.py

# 2. Chat app (terminal 2)
uv run uvicorn app:app --port 8080

Open http://localhost:8080 and ask for a chart of vehicles by propulsion type — the model calls the MCP tool and the chart renders in the conversation.

Screenshot