OpenAI pricing — relevant tiers
Per million tokens (public OpenAI rates as of 2026):
- GPT-4o: $2.50 input / $10 output
- GPT-4o mini: $0.15 input / $0.60 output
- o3: $15 input / $60 output (reasoning + tool use)
- o3-mini: $1.10 input / $4.40 output
A 15K-token tool response under GPT-4o costs $0.15. Under o3 it's $0.90. The model choice matters as much as the tool choice.
Which clients support MCP + OpenAI
Today, the practical entry points for OpenAI + MCP are:
- OpenAI Agents SDK — the official Python/Node SDK supports MCP servers as tools
- Custom GPTs — Actions can call MCP-style endpoints; limited to HTTP transport
- Continue.dev — IDE assistant that supports MCP with any provider
- Cline — VS Code extension that works with OpenAI keys
- Goose (Block) — agent framework that wraps MCP for any provider
Where MCPSpend fits
We're a transport-layer proxy. We don't know — and don't care — which LLM provider you use. We measure every MCP tool call regardless, and use provider-published rates to compute cost. As of v0.7.1 the cost engine knows: claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5, gpt-4o, gpt-4o-mini, o3, o3-mini, Gemini 2.0 Pro/Flash, Grok 2, DeepSeek Chat/Reasoner, Mistral Large/Small, Llama 3.1 405B/70B/8B, Llama 3.3 70B — and falls back to a Sonnet-equivalent baseline for anything else.
Full live list is at /api/public/pricing-models — public, no auth, refreshed monthly.
Install
npx --yes @mcpspend/proxy@latest init --key mcps_live_xxx
Auto-detects Cline, Continue.dev, Goose, Cursor, Windsurf, VS Code, Claude Desktop, and Claude Code. Wraps every MCP server it finds. The OpenAI Agents SDK isn't auto-detected (no canonical config path) — you wrap individual servers manually:
from openai import OpenAI
from agents.mcp import StdioServer
server = StdioServer(
command="npx",
args=["-y", "@mcpspend/proxy@latest", "wrap",
"--key", "mcps_live_xxx",
"--", "npx", "-y", "@modelcontextprotocol/server-filesystem", "/data"],
)Same pattern as a stdio wrap for any other client — your MCP server runs inside our subprocess.
Get a key
Free tier: 25,000 tool calls/month — covers a single heavy ChatGPT agent setup. No card. Sign up →