MCP Voice Client

Barebones MCP server + a web voice client. Tap a mic button in your browser, speak a command, and see the raw MCP tool response.

Voice → Web Speech API → transcript → OpenAI (function calling)
      → picks MCP tool + args → MCP server → response → UI

Structure

server/main.py          # FastMCP server: add, echo tools + greeting resource
client/main.py          # CLI client (original)
client/web.py           # FastAPI web client backend
client/static/index.html  # Voice UI (mic button + response cards)
pyproject.toml
.env.example

Setup

uv sync
cp .env.example .env
# edit .env and paste a fresh OPENAI_API_KEY

Run

Terminal 1 — MCP server:

uv run python -m server.main

Terminal 2 — web voice client:

uv run python -m client.web

Open http://127.0.0.1:5173 in Chrome (Web Speech API requires Chromium-based browsers). Tap the mic, say a command:

• "add two and three"
• "echo hello world"

Response card shows: transcript, tool chosen, arguments, and the raw MCP response.

CLI variant (no UI)

uv run python -m client.main

Security note

.env is gitignored. Never commit or paste your API key in chat — rotate it at https://platform.openai.com/api-keys if you do.

Extend

• Add a tool: decorate a function with @mcp.tool() in server/main.py. The web UI picks it up automatically on the next request (tools are listed fresh each call).
• Swap the LLM model: change gpt-4o-mini in client/web.py.