Authenticate
Complete GitHub device sign-in with the account that owns your Copilot subscription.
Copilot meets your coding agents.
Use GitHub Copilot from Claude Code, Codex CLI, and compatible API clients through one compatibility bridge.
OpenAI Responses
/v1/responses
Anthropic Messages
/v1/messages
Chat Completions
/v1/chat/completions
Embeddings
/v1/embeddings
Claude Code in one command.
Run the proxy, sign in to GitHub, choose your models, and paste the generated launch command. Claude Code talks to Copilot through the Anthropic Messages API.
npx copilot-proxy-api@latest start --claude-codeRequires Node.js and an active GitHub Copilot subscription.
Choose models
Select a primary model and a smaller model for Claude Code background tasks.
Launch
Paste the generated command into a new terminal and work as usual.
Codex CLI in one config block.
Codex reaches Copilot through the OpenAI Responses API. Point it at the proxy with a custom model provider, then run Codex as usual.
model = "gpt-5.5"
model_provider = "copilot_proxy"
model_reasoning_effort = "high"
[model_providers.copilot_proxy]
name = "GitHub Copilot via copilot-proxy-api"
base_url = "http://localhost:4141/v1"
wire_api = "responses"
Setting OPENAI_BASE_URL alone is not enough — recent
Codex versions read provider settings from
config.toml.
Start the proxy
Run npx copilot-proxy-api@latest start and complete
GitHub sign-in.
Pick a model
Any id your plan advertises at
/v1/models works. Check the list before pinning
one.
Run Codex
Launch codex in a second terminal. The banner
should read provider: copilot_proxy.
Familiar clients. Compatible routes.
Keep your preferred coding interface while the proxy handles translation and model routing.
Claude Code
Anthropic Messages API
POST /v1/messages
Setup guide
Codex CLI
OpenAI Responses API
POST /v1/responses
Setup guide
API clients
OpenAI-compatible and Anthropic-compatible tools
http://localhost:4141
View endpoints
Built for long agent sessions.
Compatibility is more than matching a route name. The proxy handles the edge cases coding agents depend on.
Two API shapes, one process
Serve OpenAI Responses and Chat Completions alongside Anthropic Messages from the same endpoint.
/v1/responses
/v1/messages
Streaming resilience
Keepalives and structured stream cleanup help long-running agent responses finish cleanly.
Context management
Model-aware pruning keeps oversized tool output, images, and message history within upstream limits.
Rich content translation
Translate tool calls, images, PDFs, and structured errors between client and Copilot formats.
Usage controls
Inspect quota data, pace requests, wait on rate limits, or approve requests manually.
Open usage dashboardUnofficial by design. Responsible by default.
This reverse-engineered project is not supported by GitHub and may break as upstream APIs change. Keep it local, avoid excessive automation, and follow GitHub's policies and Copilot terms.