ocp

ocp is an OpenAI-compatible HTTP endpoint backed by the GitHub Copilot SDK. It lets any client that speaks the OpenAI Chat Completions API — including Grok — route requests through your GitHub Copilot subscription instead of a separate provider key.

Under the hood, ocp starts a local ASP.NET server and proxies chat to Copilot sessions. Model discovery comes from the SDK at runtime, so the available models match what your Copilot plan and CLI expose.

Prerequisites

• A GitHub Copilot subscription
• Authenticated Copilot CLI credentials (copilot login, or GH_TOKEN / GITHUB_TOKEN / COPILOT_GITHUB_TOKEN in the environment)

Install & Run

dotnet tool install -g ocp
ocp

Run without installing

dnx -y ocp

On startup, ocp prints the working directory, the models Copilot exposes, and the listen URL (default http://localhost:11434):

ocp: using working directory: C:\Users\you\.ocp
ocp: Copilot client started.
ocp: models: gpt-5.5, gpt-5-mini, claude-sonnet-4.6, gpt-5.3-codex
ocp: listening on http://localhost:11434 (OpenAI compatible)

Options:

Flag	Description
--cwd <path>	Copilot working directory (defaults to ~/.ocp)
--help	Show usage

Install from source:

dotnet pack src/ocp/ocp.csproj
dotnet tool install -g --add-source ./bin ocp

API

Method	Path	Description
GET	/v1/models	OpenAI-compatible model list from Copilot
POST	/v1/chat/completions	Chat completion (supports stream: true)
GET	/	Health check

Example:

curl http://localhost:11434/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.5","messages":[{"role":"user","content":"Hello"}]}'

The model field must be a Copilot model id returned by /v1/models. ocp does not require an API key, though clients may send one anyway.

Configure Grok ([model.*] in config.toml)

Add entries under ~/.grok/config.toml. Each [model.<picker-name>] block tells Grok how to reach ocp; the model field is the Copilot model id sent in chat requests.

1. Start ocp and note the port from its startup log (11434 by default).

2. Add a shared base URL and one section per Copilot model you want in the picker:

[models]
default = "copilot-gpt-5.5"

# OpenAI models via Copilot
[model.copilot-gpt-5.5]
model = "gpt-5.5"
base_url = "http://localhost:11434/v1"
name = "GPT-5.5 (Copilot)"
description = "GitHub Copilot — OpenAI GPT-5.5"
context_window = 200000

[model.copilot-gpt-5-mini]
model = "gpt-5-mini"
base_url = "http://localhost:11434/v1"
name = "GPT-5 mini (Copilot)"
description = "Fast, lower-cost Copilot model"

[model.copilot-gpt-5.3-codex]
model = "gpt-5.3-codex"
base_url = "http://localhost:11434/v1"
name = "GPT-5.3 Codex (Copilot)"
description = "Code-focused Copilot model"

[model.copilot-gpt-5.4]
model = "gpt-5.4"
base_url = "http://localhost:11434/v1"
name = "GPT-5.4 (Copilot)"

# Anthropic models via Copilot
[model.copilot-claude-sonnet-4.6]
model = "claude-sonnet-4.6"
base_url = "http://localhost:11434/v1"
name = "Claude Sonnet 4.6 (Copilot)"
context_window = 200000

[model.copilot-claude-opus-4.6]
model = "claude-opus-4.6"
base_url = "http://localhost:11434/v1"
name = "Claude Opus 4.6 (Copilot)"
context_window = 200000

# Google models via Copilot
[model.copilot-gemini-2.5-pro]
model = "gemini-2.5-pro"
base_url = "http://localhost:11434/v1"
name = "Gemini 2.5 Pro (Copilot)"

3. Use the models in Grok:

grok models                  # lists custom entries alongside built-ins
/model copilot-gpt-5.5       # switch in the TUI
grok -p "refactor this" -m copilot-gpt-5.5

Notes

• Model ids — Use the exact ids from ocp's startup line or GET /v1/models. Availability depends on your Copilot plan; see GitHub's supported models.
• Section name vs model id — [model.copilot-gpt-5.5] is Grok's picker id; model = "gpt-5.5" is what ocp forwards to Copilot. They can differ, but keeping them aligned is easier to reason about.
• Port conflicts — The default port 11434 is also used by Ollama. If both run locally, ocp tries successive ports automatically; update base_url to match the URL ocp prints.
• api_backend — Omit it (defaults to chat_completions), which matches ocp's /v1/chat/completions endpoint.
• No API key — ocp has no auth middleware. You do not need api_key or env_key unless another proxy sits in front.

---