Use it with your tools.

§ 02Codex CLI

OpenAI's CLI - the real one, open-source on GitHub. Put our provider block in ~/.codex/config.toml, then create a separate ~/.codex/myco.config.toml profile file for the model settings. Codex talks to our /v1/responses endpoint.

[model_providers.myco]
name = "MyCoAI.dev"
base_url = "https://mycoai.dev/v1"
wire_api = "responses"
env_key = "MYCO_API_KEY"

model = "gpt-5.5"
model_provider = "myco"

Codex 0.134.0+ no longer reads legacy [profiles.*] entries from config.toml. Keep shared provider definitions in ~/.codex/config.toml, put model + model_provider in ~/.codex/myco.config.toml, then launch with codex --profile myco.

Then export your key and launch that profile:

export MYCO_API_KEY="sk-myco-..."
codex --profile myco

Add the export line to ~/.zshrc, ~/.bashrc, or PowerShell $PROFILE to keep the key set across sessions. Keep using codex --profile myco unless you set a separate default elsewhere.

Using the Codex desktop app? GUI processes don't inherit shell exports. Set the API key env var (the one named in env_key above) as a persistent OS variable - on Windows: System Properties > Environment Variables; on macOS: launchctl setenv or your login profile - then restart the app. If you use a named profile there too, keep the same ~/.codex/myco.config.toml file.

§ 03Claude Code

Anthropic's CLI. Three ways to point it at us — pick one. Global routes everything through us, per-project keeps your other repos on your normal account, shell is a one-shot.

Global (all projects)

Same JSON, just placed at the user-level config so every project on this machine picks it up automatically.

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://mycoai.dev",
    "ANTHROPIC_AUTH_TOKEN": "sk-myco-...",
    "ANTHROPIC_MODEL": "claude-sonnet-4-6",
    "ANTHROPIC_SMALL_FAST_MODEL": "claude-haiku-4-5-20251001",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
  }
}

Path: ~/.claude/settings.json on macOS and Linux · %USERPROFILE%\.claude\settings.json on Windows.

Per-project

Drop this in the repo where you want to use our gateway. Other projects stay on your normal Claude account.

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://mycoai.dev",
    "ANTHROPIC_AUTH_TOKEN": "sk-myco-...",
    "ANTHROPIC_MODEL": "claude-sonnet-4-6",
    "ANTHROPIC_SMALL_FAST_MODEL": "claude-haiku-4-5-20251001",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
  }
}

Shell environment

Set the variables in your shell — affects just the current terminal.

export ANTHROPIC_BASE_URL="https://mycoai.dev"
export ANTHROPIC_AUTH_TOKEN="sk-myco-..."
claude

To make it stick across sessions, paste those export lines into your shell profile: ~/.zshrc, ~/.bashrc, or PowerShell $PROFILE.

ANTHROPIC_MODEL and ANTHROPIC_SMALL_FAST_MODEL pick which models Claude Code routes to — use any Claude model from your catalog.

§ 04OpenCode

Cross-platform code agent built on the Vercel AI SDK. Drop us in as a custom provider in your global opencode.json — every project on the machine picks it up.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "myco": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "MyCoAI.dev",
      "options": {
        "baseURL": "https://mycoai.dev/v1",
        "apiKey": "sk-myco-..."
      },
      "models": {
        "gpt-5.5": {}
      }
    }
  },
  "model": "myco/gpt-5.5"
}

Path: ~/.config/opencode/opencode.json on macOS and Linux · %USERPROFILE%\.config\opencode\opencode.json on Windows.

Then run it from anywhere:

opencode run --model myco/gpt-5.5 "Hello"

§ 05OpenAI SDK

Set base_url — every existing call keeps working unchanged.

from openai import OpenAI

client = OpenAI(
    api_key="sk-myco-...",
    base_url="https://mycoai.dev/v1",
)

resp = client.chat.completions.create(
    model="gpt-5.5",
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)

§ 06Anthropic SDK

Use this section only for clients you have tested on /v1/messages. Current Claude models are intended for Claude Code, not generic raw Anthropic SDK calls.

§ 07Image generation

Generate PNGs from a text prompt via the OpenAI-compatible /v1/images/generations endpoint. Works with the standard OpenAI SDK — just set base_url.

curl https://mycoai.dev/v1/images/generations \
  -H "Authorization: Bearer sk-myco-..." \
  -H "Content-Type: application/json" \
  --max-time 180 \
  -d '{
    "model": "gpt-image-2",
    "prompt": "a friendly orange cat sitting on a desk, illustration",
    "n": 1,
    "size": "1024x1024"
  }'

from openai import OpenAI
import base64

client = OpenAI(
    api_key="sk-myco-...",
    base_url="https://mycoai.dev/v1",
    timeout=180,  # generation takes 60-80s
)

resp = client.images.generate(
    model="gpt-image-2",
    prompt="a friendly orange cat sitting on a desk, illustration",
    n=1,
    size="1024x1024",
)

with open("out.png", "wb") as f:
    f.write(base64.b64decode(resp.data[0].b64_json))

Editing an existing image

Send a PNG plus a prompt to /v1/images/edits and gpt-image-2 returns an edited PNG. Same auth and SDK as generation — just use client.images.edit() and pass the source file. The endpoint is multipart/form-data (not JSON).

curl https://mycoai.dev/v1/images/edits \
  -H "Authorization: Bearer sk-myco-..." \
  --max-time 180 \
  -F "model=gpt-image-2" \
  -F "[email protected]" \
  -F "[email protected]" \
  -F "prompt=turn the sky blue and add a rainbow" \
  -F "n=1" \
  -F "size=1024x1024"

from openai import OpenAI
import base64

client = OpenAI(
    api_key="sk-myco-...",
    base_url="https://mycoai.dev/v1",
    timeout=180,  # edits take 20-40s
)

with open("source.png", "rb") as img, open("mask.png", "rb") as mask:
    resp = client.images.edit(
        model="gpt-image-2",
        image=img,
        mask=mask,                # optional — omit to edit the whole image
        prompt="turn the sky blue and add a rainbow",
        n=1,
        size="1024x1024",
    )

with open("edited.png", "wb") as f:
    f.write(base64.b64decode(resp.data[0].b64_json))

§ 08Cline & Roo Code

VS Code extensions. In the provider dropdown choose OpenAI Compatible, then fill in these fields:

Provider

OpenAI Compatible

Base URL

https://mycoai.dev/v1

API Key

sk-myco-...

Model ID

gpt-5.5

§ 09Aider

Aider reads OpenAI-style environment variables. Point them at us and pick a model:

export OPENAI_API_BASE="https://mycoai.dev/v1"
export OPENAI_API_KEY="sk-myco-..."
aider --model openai/gpt-5.5

§ 10Any other client

Anything that lets you set a base URL and an API key works. Use the OpenAI surface for most tools, the Anthropic surface for Claude-native ones.

§ 11Troubleshooting

• 401 invalid_api_key
  Check the key is active in the dashboard and was copied whole, including the sk- prefix.
• 402 insufficient_balance
  Top up your wallet — your balance is below the request's estimated cost.
• 404 model_not_found
  The model is not in your catalog. Call GET /v1/models to see what is available to you.
• slow first request
  The first call to a cold channel can lag. We fail over automatically when a channel is down.