YA

ClaudeBridge

36 views

Project Links

How much is your Claude Pro subscription actually worth?

I was into my second week of my $20/month subscription before I realized I had no idea. The official client shows me nothing. No token counts, no cost estimates. Just a vague "you've used X% of your limit" that resets at the end of the window.

I tried to use Claude Code across 2 machines and it would make tracking the usage by tools like ccusage impossible.

That's when I started wondering: am I getting my money's worth? Or was I paying $20 to guess?

The Problem With Flying Blind

Here's what happens when you use Claude Pro the "normal" way:

  • You send requests until you hit a wall
  • The wall appears randomly mid-conversation
  • You have no idea what 10% or 90% of your usage means in terms of token or money.
  • You definitely don't know what those requests cost in API terms

It's like driving without a speedometer.

What ClaudeBridge Actually Does

ClaudeBridge is a self-hosted gateway that sits between your tools and Claude Pro. It exposes both OpenAI-compatible and Anthropic-compatible endpoints, which means you can point virtually any LLM client at it.

and that's oppurtunity to cram a lot of features in the middle every request gets tracked. Tokens in, tokens out, cost estimates, session boundaries, external rate limit status, internal-per-application rate-limits as well. You get real-time dashboards showing exactly where you stand in your 5-hour and 7-day usage windows.

Setting It Up

Run it as a Python module or Docker container, go through the OAuth flow, and you're basically done. The web UI gives you everything you need for basic monitoring.

For the full experience with Prometheus metrics:

  1. Deploy it somewhere persistent (I run mine on Kubernetes with a Tailscale sidecar)
  2. Point your clients at your bridge URL
  3. Set your API key from the web UI
  4. Watch the dashboards populate

One heads up: the usage tracking can get wonky if you're using your subscription both with and without the bridge simultaneously. Pick one path and stick to it.

Two Ways To Use It (And The ToS Question)

The "Probably Fine" Way

If you're using Claude Code and just want better visibility into your usage, you're in the clear. You're still using the official client—the bridge is just a proxy that adds observability.

Set it up like this:

alias claude='ANTHROPIC_BASE_URL="your-bridge-machine" ANTHROPIC_API_KEY="your-key" claude'

Now you get all the tracking without changing your workflow.

The "Your Call" Way

You can use ClaudeBridge with non-Anthropic tools (OpenAI's Codex or OpenCode). This opens up your subscription to multiple users and applications.

Does this violate the ToS? Oh yeah, probably.

The Real Value

So how much is that $20 subscription actually worth?

Based on my usage over the past three months: about $143 in equivalent API costs.

That's the difference between guessing and knowing. Between hitting mystery limits and planning your usage. Between using Claude on one machine and using it wherever you actually work.

The bridge has been rock-solid for me for 3+ months. It's open source, self-hosted, and puts you back in control of a service you're already paying for.

It also answer clearly a simple questions: Should a model compete with Sonnet/Opus, how much cheaper should its subscription be to be worth the switch?
(that's assuming that every model uses the same amount of token to reason through the same task and right now GLM 4.7 is teaching me that it really isn't the case)