Keep MCP tool loops visible and capped.
MCP adds tools; ProxyLLM carries the model calls around them. Give each host a scoped key and the loop runs through Codex Hosted on your ChatGPT subscription, with request logs and a spending limit.
$129/month SaaS. Bring your own model keys. No inference markup.
Three steps to connect.
Create a host key
Generate a ProxyLLM key for the MCP host that will call models after tool results come back.
Set OpenAI-compatible env vars
Point the host or model adapter at https://api.proxyllm.ai/v1 and pass the ProxyLLM key as the API key.
Run loops on your subscription
MCP sessions create many small reasoning calls around tools. Codex Hosted bills them to your ChatGPT plan, your API key handles overflow past a plan limit, and every request lands in the log.
Configure the host's model endpoint.
Use OpenAI-compatible environment variables where your MCP host supports them.
{
"env": {
"OPENAI_BASE_URL": "https://api.proxyllm.ai/v1",
"OPENAI_API_KEY": "pk_live_..."
}
}Run your AI workloads on your ChatGPT subscription.
ProxyLLM runs OpenAI's Codex for you, signed in with your own ChatGPT account. Your apps call one OpenAI-compatible endpoint and the work bills to your flat plan instead of per-token API pricing.
Tools make calls multiply.
Every tool result triggers another model call. Through Codex Hosted those calls bill flat to your ChatGPT subscription instead of stacking per-token charges.