v2.0.0-rc1

ccproxy v2.0.0-rc1

Recent shifts in the landscape necessitate change: ccproxy is no longer a LiteLLM addon, or even just a proxy. It is no longer catered entirely to Claude Code, and can now serve as a universal inspector for any AI application that crosses your network.

LiteLLM has been decapitated. It no longer runs as the executing process or starts the server; it has been demoted to an import, used solely for its cross-provider translation layer. The recent supply chain attack still motivates complete removal, but I decided against it for now to avoid breaking applications that still expect the API spec LiteLLM provided.

The software itself is production-ready. What remains is largely its presentation, and more importantly for you, the readme and documentation. What is currently provided is sufficient but mostly machine-generated; I have at least reviewed it and it is complete.

Suggestions or feedback welcome on the PR.

Full Changelog: https://github.com/starbaser/ccproxy/commits/v2.0.0-rc1

v1.2.0 - The Escape Plan

This release is a step in a new direction I hope to take this project: Jailbreaking Claude Code. I see ccproxy as the bedrock for creating a near replica of the Claude Code harness, an escape hatch for bringing its capabilities to any command-line tool/harness. Currently, experimenting with a forked version of https://github.com/charmbracelet/crush/ is showing much promise. If you would like to learn more, please join my new discord.

---

Features

Keyless API Access

Multi-provider OAuth support via oat_sources configuration. When enabled, the proxy automatically forwards the correct OAuth token based on routing, eliminating the need to manage API keys manually. Includes custom User-Agent support per token source.

Improved Hooks

Added hooks that accept parameters. This is compatible with inline hooks and is backwards compatible. See the ccproxy.yaml example file for reference.

LangFuse Integration

First-class LangFuse observability support. Claude Code sessions are automatically extracted and grouped as LangFuse sessions, with HTTP headers captured as trace metadata (sensitive values redacted). See .env.example for setting up your LangFuse credentials.

CLI Improvements

The ccproxy status command output has been improved to display hooks and model deployments. The handler file (ccproxy.py) is now auto-generated on startup, with detection to preserve custom handler files.

Bug Fixes

• Use bundled litellm from venv instead of PATH (fixes "No module named 'backoff'" error)

Documentation

• Add installation and troubleshooting guides
• Fix installation commands and prioritize PyPI
• Document auto-generated handler workflow

Chores

• Update Opus model references to claude-opus-4-5-20251101
• Update ccproxy.yaml template to current format (oat_sources, capture_headers)
• Comprehensive test coverage improvements