MrVinci OpenAI API

One host, landing and API together

The marketing pages you're reading and the API share this domain — the browser gets HTML at /, your client hits the API path below.

What you need

• Node.js 18+ (for the npx installer)
• A MrVinci key — sk_mv_… — from an admin
• A supported client: Claude Code, Cursor, Cline, or any OpenAI SDK

The one-minute path

The npx mrvinciapi wizard asks for your key, lets you pick your client, writes the config, and verifies the connection. Zero dependencies, pure Node — it only edits local files, never sends anything anywhere.

npx mrvinciapi --openai

# Claude Code
npx mrvinciapi sk_mv_YOUR_KEY
# Codex (OpenAI Responses)
npx mrvinciapi sk_mv_YOUR_KEY --codex
# Cursor / Cline
npx mrvinciapi sk_mv_YOUR_KEY --openai

# print env, change nothing
npx mrvinciapi sk_mv_YOUR_KEY --print
# revert Claude Code
npx mrvinciapi --undo

Codex, via the Responses API

Codex defaults to wire_api = "responses". We serve a full OpenAI Responses endpoint at /v1/responses (streaming + non-streaming), so Codex works out of the box — npx mrvinciapi <key> --codex writes this for you.

Or add it to ~/.codex/config.toml and run codex --profile mrvinci:

[model_providers.mrvinci]
name = "MrVinci"
base_url = "https://openai.mrvinci.app/v1"
wire_api = "responses"
env_key = "MRVINCI_API_KEY"

[profiles.mrvinci]
model = "gpt-5.4"
model_provider = "mrvinci"

Export the key first: export MRVINCI_API_KEY="sk_mv_…".

Configure it yourself

Claude Code

Add to ~/.claude/settings.json:

{
  "env": {
    "OPENAI_API_KEY": "sk_mv_YOUR_KEY",
    "OPENAI_BASE_URL": "https://openai.mrvinci.app"
  }
}

Cursor / Cline

Base URL https://openai.mrvinci.app, API key sk_mv_…, model DeepSeek-V4-Pro.

POST /v1/chat/completions

curl https://openai.mrvinci.app/v1/chat/completions \
  -H "Authorization: Bearer sk_mv_YOUR_KEY" \
  -H "content-type: application/json" \
  -d '{
    "model": "DeepSeek-V4-Pro",
    "messages": [{"role":"user","content":"Hello"}]
  }'

Authenticate with x-api-key: sk_mv_… or Authorization: Bearer sk_mv_…. List models at /v1/models.

POST /v1/images/generations

Send the OpenAI-compatible model, prompt, size, quality, and n fields. Image calls are billed per image, not per token.

curl https://openai.mrvinci.app/v1/images/generations \
  -H "Authorization: Bearer sk_mv_YOUR_KEY" \
  -H "content-type: application/json" \
  -d '{
    "model": "gpt-image-1-5",
    "prompt": "A brass astrolabe on a drafting table",
    "size": "1024x1024",
    "quality": "medium",
    "n": 1
  }'

Models: gpt-image-1-5, flux-2-pro. size and quality select the GPT Image price tier; FLUX is flat-rate. This generation endpoint does not accept reference images.

Tools, streaming, vision, image generation, caching

Everything the Chat Completions / Responses API supports works through the gateway — these aren't add-ons, they're the same wire format you already use.

Tool use (function calling)

Send a tools array exactly as the OpenAI SDK does — tool_use blocks stream back exactly as the API specifies. Agents like Claude Code, Cursor, and Cline run their whole tool loop through the gateway unmodified.

curl https://openai.mrvinci.app/v1/chat/completions \
  -H "Authorization: Bearer sk_mv_YOUR_KEY" \
  -H "content-type: application/json" \
  -d '{
    "model": "DeepSeek-V4-Pro",
    "tools": [{
      "type": "function",
      "function": {
        "name": "get_weather",
        "description": "Current weather for a city",
        "parameters": {"type":"object","properties":{"city":{"type":"string"}},"required":["city"]}
      }
    }],
    "messages": [{"role":"user","content":"Weather in Florence?"}]
  }'

Error reference

Errors come back in the OpenAI error envelope, so SDK retry logic behaves correctly. The rule: 429 and 529 are retryable (wait, then retry); 4xx everything else means fix the request.

When something's off