Does using a proxy with Cursor add latency?

Stockyard's middleware chain adds approximately 400 nanoseconds. Cursor's API calls take 1-30 seconds. The proxy overhead is unmeasurable in practice.

Can I use Cursor with multiple providers through Stockyard?

Yes. Set API keys for OpenAI, Anthropic, and others. Use model aliasing to route Cursor requests to whichever provider you want without changing Cursor's configuration.

Will Cursor's features still work through a proxy?

Yes. Stockyard is fully OpenAI-compatible. Tab completions, chat, composer, and all Cursor features work exactly as before. The only difference is requests route through Stockyard first.

Stockyard LLM Proxy for Cursor — Track Costs, Cache Requests, Add Failover

The problem with Cursor and API costs

Cursor sends a lot of LLM requests. Tab completions, inline edits, chat, composer. Each one costs money, and Cursor does not tell you how much. You find out when the OpenAI bill arrives.

A proxy sits between Cursor and the API. Every request gets logged with the model, token count, and cost. You see exactly what Cursor is doing and what it costs, in real time.

Set up in 2 minutes

# 1. Install Stockyard
curl -fsSL stockyard.dev/install.sh | sh
export OPENAI_API_KEY=sk-...
stockyard

# 2. In Cursor settings, set:
OpenAI API Base: http://localhost:4200/v1
# That's it. Cursor now routes through Stockyard.

Every tab completion, every chat message, every composer edit now flows through Stockyard. You get per-request cost tracking, response caching, and all 76 middleware modules.

What you get

Cost visibility. See every request Cursor makes with the model, token count, and exact cost. Stockyard's cost dashboard shows daily breakdowns so you know exactly where your money goes.

Response caching. Cursor re-sends context when you edit a file. With prompt caching enabled, identical requests return instantly from cache. No API call, no cost. This alone can cut Cursor's API spend by 30-50% during iterative coding.

Model swapping. Use model aliasing to route Cursor's requests to a cheaper model without changing your editor settings. Point gpt-4o at deepseek-chat for routine completions, keep the real GPT-4o for complex reasoning.

Failover. When OpenAI goes down, Stockyard fails over to Anthropic or another provider. Cursor keeps working.

Also works with other editors

The same setup works with Windsurf, GitHub Copilot, Cline, and Aider. Any editor that lets you set a custom API base URL works with Stockyard. The MCP server also provides native integration for editors that support it.