Vox MCP
Route agent completions across eight cloud providers, OpenRouter, or local models without injecting a fixed system prompt into every call.
Overview
Vox MCP is a MCP server for the Build phase that exposes a multi-model LLM gateway (eight providers, OpenRouter, local, and custom OpenAI-compatible URLs) to your agent without injecting its own system prompt.
What is this MCP server?
- Stdio MCP server (vox-mcp on PyPI) for multi-provider LLM routing
- Eight named provider API keys plus optional CUSTOM_API_URL for OpenAI-compatible endpoints
- Explicit positioning: no system prompt injection from the gateway layer
- Supports local models alongside hosted providers for cost and privacy experiments
- Per-provider secrets (Gemini, OpenAI, Anthropic, xAI, DeepSeek, Moonshot, OpenRouter)
- Server version 0.3.0 on PyPI identifier vox-mcp
- Eight documented provider API key slots plus CUSTOM_API_URL
- Stdio transport only in published server manifest
What problem does it solve?
Solo builders waste time rewiring agents every time they want to try another LLM vendor or run a local model alongside cloud APIs.
Who is it for?
Indie developers who already run MCP in Claude Code or Cursor and want provider flexibility without forking agent code for each model.
Skip if: Teams that only ever use one model inside the host app with no MCP, or builders who need hosted-only SaaS with zero secret management.
What do I get? / Deliverables
After you register Vox MCP with stdio and your API keys, your agent can call one gateway to reach multiple models while you keep system prompts in your repo or skills.
- Registered stdio MCP gateway reachable from your coding agent
- Ability to invoke multiple hosted or local models through one tool surface
- Provider credentials isolated in environment variables per vendor
Recommended MCP Servers
Journey fit
Canonical shelf is Build because you wire Vox into Claude Code, Cursor, or Codex as stdio MCP during product and agent development. agent-tooling fits a gateway MCP that exposes multi-model chat as tools rather than a single app feature integration.
How it compares
MCP multi-model gateway integration, not a curated skills marketplace or a single-purpose codegen skill.
Common Questions / FAQ
Who is Vox MCP for?
Solo and indie builders using Claude Code, Cursor, Codex, or generic MCP hosts who want one stdio server to reach several LLM providers and optional local models.
When should I use Vox MCP?
Use it during Build and Operate when you are prototyping features, comparing model quality, or routing tasks to cheaper or local models without changing your agent’s core prompts.
How do I add Vox MCP to my agent?
Install the vox-mcp PyPI package, configure stdio MCP in your host, and set the provider API key environment variables (and CUSTOM_API_URL if needed) before starting the server.