Skip to main content
Glama

voice-agent-mcp

voice-agent-mcp is a voice-driven OpenAI tool-calling agent whose tools are defined once and also served to MCP clients over stdio.

Most LLM demos stop at generating text. This project focuses on the harder boundary: letting a model perform real actions without giving up type safety, validation, observability, or control. Speech is transcribed, an OpenAI Responses API agent selects typed tools, every call runs through one validated executor, and the answer is spoken back. The same tool definitions are served to Claude and other MCP clients over stdio. It is a production-grade reference implementation with documented limits, built milestone by milestone; the current state is the M0 foundation (tooling, quality gates, configuration, logging). Generated speech is AI-generated.

License: MIT. Security: SECURITY.md. Contributing: CONTRIBUTING.md.

Documentation

Need

Read

How it is built

Architecture overview

How one tool definition serves OpenAI and MCP

Tool system

Trust boundaries and action control

Security

The rules the codebase enforces

Invariants

Why the big choices were made

Decisions

How correctness is tested

Testing

Run it locally

Local setup

Related MCP server: paperclip-mcp

Quick start

npm install
npm run check
npm run dev -- --help

Related MCP Connectors

Related MCP Servers

  • A
    license
    A
    quality
    C
    maintenance
    Exposes a verified tool registry (calculator, sandboxed file read, web fetch) over MCP stdio, enabling any MCP-capable client to reuse the same tools from the inspectable ReAct loop.
    3
    MIT
  • F
    license
    C
    quality
    B
    maintenance
    Schema-driven MCP server that exposes all Paperclip API operations as typed MCP tools over stdio or Streamable HTTP, supporting both authenticated and local_mode deployments.
    100
    -