██████╗ ██████╗██████╗ 
 ██╔════╝██╔════╝██╔══██╗   Claude Code Proxy (CCP)
 ██║     ██║     ██████╔╝   by SuparvaCode
 ██║     ██║     ██╔═══╝ 
 ╚██████╗╚██████╗██║
  ╚═════╝ ╚═════╝╚═╝ 

⚡ CCP — Claude Code Proxy

Route Claude Code and Codex CLI through any AI provider — Groq, Gemini, DeepSeek, Bedrock, Ollama, and more.

CCP is a self-hosted Node.js proxy server that speaks the Anthropic and OpenAI API wire formats. It intercepts requests from Claude Code or Codex CLI and forwards them to whichever backend AI provider you configure — with streaming, tool calling, and model routing all handled transparently.

Built and maintained by Suparva.

---

What it does

Claude Code and Codex CLI are hardcoded to talk to Anthropic's API. CCP sits in between and translates:

Claude Code / Codex CLI
        │
        ▼  (Anthropic/OpenAI wire format)
  CCP Proxy  ←── Admin Panel (http://127.0.0.1:8082/admin)
        │
        ▼  (provider-native format)
  Groq / Gemini / DeepSeek / Ollama / Bedrock / ...

You point your client at http://127.0.0.1:8082 instead of Anthropic's servers, then pick any provider and model from the admin panel.

---

Features

• 20 providers out of the box — cloud and local (see table below)
• SSE streaming — full server-sent event translation so streaming responses work exactly like Anthropic's
• Tool / function calling — passes tool schemas and results through correctly
• Reasoning/thinking blocks — DeepSeek R1 and Gemini thinking tokens surfaced as Anthropic thinking blocks
• Model routing — map Claude variants (Opus / Sonnet / Haiku) to any provider model via the admin UI
• MCP Tools manager — configure Model Context Protocol servers (web search, filesystem, GitHub, etc.) and generate a ready-to-paste ~/.claude.json snippet
• Request logs — full history with token counts, latency, and status per request
• Analytics dashboard — daily charts for tokens used, request volume, and error rate
• Built-in playground — chat directly against any configured provider/model from the browser
• AES-256-CBC encryption — all API keys are encrypted at rest in the local SQLite database
• Persistent dark/light theme

---

Requirements

• Node.js v18+ — nodejs.org
• No other global dependencies required

---

Installation

🪟 Windows (PowerShell)

Open PowerShell and run:

irm -useb https://github.com/ghraw/SuparvaCode/ccp-proxy/main/install.ps1 | iex

The window will stay open throughout the install. Everything is automatic — the installer clones the repo, installs deps, builds the UI, and registers ccp-start. Installs to %USERPROFILE%\.ccp-proxy.

🍎 macOS / 🐧 Linux

curl -fsSL https://github.com/ghraw/SuparvaCode/ccp-proxy/main/install.sh | bash

Installs to ~/.ccp-proxy and registers ccp-start in ~/.local/bin. Open a new terminal after install (or run source ~/.bashrc) for ccp-start to be available.

Manual (all platforms)

git clone https://github.com/SuparvaCode/ccp-proxy.git
cd ccp-proxy
npm run install:all
cp server/.env.example server/.env
# Edit server/.env — set CCP_ENCRYPTION_SECRET (min 32 chars) and optionally CCP_AUTH_TOKEN
npm run build:admin
npm link   # registers the global ccp-start command

🗑️ Uninstalling

To completely remove the proxy, files, PATH updates, and global commands:

Windows (PowerShell)

irm -useb https://github.com/ghraw/SuparvaCode/ccp-proxy/main/uninstall.ps1 | iex

macOS / Linux (Bash)

curl -fsSL https://github.com/ghraw/SuparvaCode/ccp-proxy/main/uninstall.sh | bash

---

Starting the server

After installation, run from any terminal:

ccp-start

Or use the local launcher in the project folder:

Windows:      .\ccp.cmd
macOS/Linux:  ./ccp

The server starts on port 8082 by default (configurable in the admin panel under Settings → Server Port).

⚡ CCP Proxy running at http://127.0.0.1:8082
📋 Admin panel:      http://127.0.0.1:8082/admin
🔌 Anthropic API:    http://127.0.0.1:8082/v1/messages
🤖 OpenAI API:       http://127.0.0.1:8082/v1/responses

---

Connecting your client

Claude Code

export ANTHROPIC_BASE_URL=http://127.0.0.1:8082
export ANTHROPIC_AUTH_TOKEN=super
claude

On Windows (PowerShell):

$env:ANTHROPIC_BASE_URL = "http://127.0.0.1:8082"
$env:ANTHROPIC_AUTH_TOKEN = "super"
claude

Codex CLI

export CCP_CODEX_API_KEY=super

codex \
  -c model_provider="ccp" \
  -c model_providers.ccp.name="Claude Code Proxy" \
  -c model_providers.ccp.base_url="http://127.0.0.1:8082/v1" \
  -c model_providers.ccp.env_key="CCP_CODEX_API_KEY" \
  -c model_providers.ccp.wire_api="responses"

The default auth token is super. Change it in Admin → Settings and update your env var to match.

---

Selecting a model

Address any model using the {provider_id}/{model_id} format:

google/gemini-2.5-pro
groq/llama-3.3-70b-versatile
deepseek/deepseek-chat
bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0
ollama/llama3.2

Or set up routing rules in Admin → Model Routing to automatically map Claude variants (e.g. claude-3-5-sonnet) to a provider/model of your choice — so you never have to change your client config.

---

Admin panel

Open http://127.0.0.1:8082/admin after starting the server.

Page	What it does
Dashboard	Token usage, request count, error rate, active providers at a glance
Providers	Configure API keys and base URLs for each provider
Models	Browse and cache available models per provider
Model Routing	Map Claude variant names to specific provider/model pairs
Playground	Chat interface to test any provider/model directly
Request Logs	Full log of every proxied request with tokens and latency
Analytics	Daily charts for usage over time
MCP Tools	Configure MCP servers for Claude Code (web search, filesystem, GitHub…)
Settings	Change auth token, port, log level, and server restart

---

Provider registry

Provider	ID	Type	Notes
Google AI Studio	google	Cloud	Gemini models
DeepSeek	deepseek	Cloud	Includes R1 reasoning
OpenRouter	openrouter	Cloud	Aggregates 200+ models
Groq	groq	Cloud	Ultra-fast inference
Mistral AI	mistral	Cloud
Mistral Codestral	codestral	Cloud	Code-optimised
Cerebras	cerebras	Cloud
Fireworks AI	fireworks	Cloud
NVIDIA NIM	nvidia	Cloud
Together AI	together	Cloud
xAI (Grok)	xai	Cloud
Cohere	cohere	Cloud
Amazon Bedrock	bedrock	Cloud	IAM credentials or gateway
Kimi (Moonshot)	moonshot	Cloud	OpenAI-compatible endpoint
GLM (Zhipu AI)	zhipu	Cloud	OpenAI-compatible endpoint
MiniMax	minimax	Cloud	OpenAI-compatible endpoint
01.AI (Yi)	yi	Cloud	OpenAI-compatible endpoint
Ollama	ollama	Local	Runs on localhost:11434
LM Studio	lmstudio	Local	Runs on localhost:1234
llama.cpp	llamacpp	Local	Runs on localhost:8080

---

API endpoints

Endpoint	Description
POST /v1/messages	Anthropic Messages API (Claude Code)
POST /v1/messages/count_tokens	Token counting
GET /v1/models	List available models
POST /v1/responses	OpenAI Responses API (Codex CLI)
GET /health	Health check
GET /admin/*	Admin panel UI
GET/PUT /api/admin/*	Admin REST API

---

Configuration

Server configuration is stored in a local SQLite database (server/data/ccp.db) and managed through the admin panel. For production deployments, set these in server/.env:

Variable	Default	Description
CCP_ENCRYPTION_SECRET	insecure default	Change this. Min 32 chars. Used to encrypt API keys at rest.
CCP_AUTH_TOKEN	super	Bearer token your clients send. Set to something strong.
PORT	8082	Override the server port (or change via admin Settings)
HOST	127.0.0.1	Bind address. Use 0.0.0.0 to expose on the network.

---

MCP Tools

CCP includes an MCP Tools manager (Admin → MCP Tools) that lets you configure Model Context Protocol servers and generate a ~/.claude.json snippet for Claude Code.

Built-in templates: Brave Search, Filesystem, Puppeteer Browser, HTTP Fetch, GitHub, PostgreSQL, Memory/Knowledge Graph, Custom SSE.

Click Get Config in the MCP Tools page to copy the ready-to-paste config block.

---

License

MIT — see LICENSE.