SuperCompress Proxy — One Command, Every Coding Agent, ~65% Fewer Tokens

If you use a coding agent like Cursor or Windsurf every day, you've probably noticed two things:

1. Your token usage is massive — every file context, conversation history, and tool output gets sent to the LLM as prompt tokens
2. Your bill grows with it — at GPT-4 Turbo pricing ($10/1M input tokens), a heavy Cursor user spends $50–200+/month on prompt tokens alone

We built SuperCompress to solve this: it compresses prompt context before it reaches the LLM, cutting token costs by ~65% with no quality loss. The problem was always integration — adding middleware, configuring agents, setting up proxies. Too many steps.

Not anymore.

One Command

That's it. npm install alone does not rewrite agent configs. setup links your SuperCompress account, then:

1. Detects coding agents installed on your machine
2. Installs the compress_context MCP plugin for supported agents
3. Optionally opts into a localhost OpenAI/Anthropic base-URL proxy if you pass --proxy

From this moment forward, agents can call compress_context on large dumps before they burn tokens. Provider-managed subscription sessions stay on MCP — no API-key mode required.

Works With Compatible Agents

When a new agent exposes a configurable OpenAI- or Anthropic-compatible endpoint, run supercompress setup again to detect it and follow its manual configuration steps.

How It Works

The proxy runs on your machine at localhost:8080. Your coding agents are configured to send their LLM requests there instead of directly to OpenAI/Anthropic. The proxy:

1. Receives the messages from your coding agent
2. Compresses the context using the SuperCompress API (billable through your plan)
3. Forwards the compressed messages to the real provider (OpenAI/Anthropic)
4. Returns the response back to your coding agent

Your provider API key stays on your machine. Only the context text is sent to the SuperCompress API for compression. The proxy adds ~60ms of overhead — invisible compared to a 5–30 second LLM call.

The Cost Math

Metric	Without Proxy	With Proxy
Tokens per request	~4,000	~1,400
Cost per 1K requests (GPT-4 Turbo)	$40	~$14
Monthly cost (heavy Cursor user)	$50–200	~$18–70
Context window fit	100K window fills fast	~3x more effective space

What About SuperCompress Pricing?

Every compression call goes through the SuperCompress API. Free accounts get 5M tokens/month; enable pay-as-you-go for more:

Tier	Price	Tokens	With a Coding Agent
Free	$0	5M / month included	Light daily use for most solo agents
Pay as you go	$1 / 1M tokens	Unlimited after free	Heavy Cursor / Claude Code days

The math works in your favor. If you're spending $50–200/month on Cursor tokens, saving 65% means you save $33–130/month. Over the free 5M, SuperCompress overage is $1 per million tokens compressed — a small fraction of what you save on the model bill.

For Teams

Run the proxy as a shared Docker container for your whole team:

docker run -p 8080:8080 \
  -e SUPERCOMPRESS_API_KEY=sc_live_team_key \
  -e PROXY_PORT=8080 \
  supercompress/proxy

Every developer on the team configures their agent to point at the shared proxy. Usage is centralized under one API key with pay-as-you-go enabled. The dashboard shows per-developer usage.

What Developers Are Saying

"I was spending $180/month on Cursor. After installing the SuperCompress proxy, it dropped to ~$60/month. The setup took 30 seconds and I haven't touched it since."

"The auto-detection found all three of my agents (Cursor, Continue, Claude Code) and configured them all at once. This is how every developer tool should work."

Get Started

You'll need a SuperCompress account. If you don't have one, setup opens supercompress.dev/dashboard to sign in and link this install.

For complete documentation, see the coding agents guide. Source: GitLab.