Proxy Station
サインアップ

お使いのツールを接続

すでに使っているエディタや CLI をこのエンドポイントに向けるだけです。アプリのインストールは不要 —— base URL とキー、そして動作を証明する 1 コマンドだけ。

base URL は 2 種類、キーは 1 つ

OpenAI-compatiblethe client appends /chat/completions
https://proxystation.co/v1
Anthropic-compatiblethe client appends /v1/messages itself
https://proxystation.co

この違いは見た目の問題ではありません。OpenAI 形式のクライアントは与えられた URL に /chat/completions を付け足すのでバージョン付きのパスが必要です。Claude Code と Anthropic SDK は /v1/messages を自分で付け足すので裸のオリジンが必要です。逆にすると /v1/v1/messages となり 404 が返りますが、多くのクライアントはこれをパスの誤りではなく接続失敗として報告します。

ターミナル

Claude Code

https://proxystation.co

インストール

npm install -g @anthropic-ai/claude-code

設定~/.claude/settings.json

export ANTHROPIC_BASE_URL=https://proxystation.co
export ANTHROPIC_AUTH_TOKEN=sk-XXXXXXXXXXXXXXXX

# Persist it in ~/.claude/settings.json instead of your shell profile:
{
  "env": {
    "ANTHROPIC_BASE_URL": "https://proxystation.co",
    "ANTHROPIC_AUTH_TOKEN": "sk-XXXXXXXXXXXXXXXX"
  }
}

開く前に検証

curl -X POST "https://proxystation.co/v1/messages" \
  -H "Authorization: Bearer sk-XXXXXXXXXXXXXXXX" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-sonnet-5","max_tokens":8,"messages":[{"role":"user","content":"ok"}]}'

期待される応答： A body starting {"id":"msg_… — the URL and the credential both work.

  • No /v1 in the base URL. Claude Code appends /v1/messages itself; adding it yields /v1/v1/messages, which this endpoint answers 404.
  • ANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEY both work here — tested. Prefer AUTH_TOKEN: API_KEY triggers a one-time approval prompt in interactive sessions, and if it was ever declined the key is ignored with no error.
  • A value in ~/.claude/settings.json BEATS a shell export, which is the opposite of what most people expect. A stale entry there will quietly override a fresh export.
  • Setting only the base URL does not switch billing. Without a credential variable a saved claude.ai login stays active — run /status inside the CLI and check it names an auth token, not a login method.

Codex CLI

https://proxystation.co/v1

インストール

npm install -g @openai/codex

設定~/.codex/config.toml

# 1. Put the key in your shell, under the name config.toml will reference:
export PROXYSTATION_API_KEY=sk-XXXXXXXXXXXXXXXX

# 2. ~/.codex/config.toml
model_provider = "proxystation"
model = "gpt-5.6-luna"

[model_providers.proxystation]
name = "proxystation"
base_url = "https://proxystation.co/v1"
env_key = "PROXYSTATION_API_KEY"
wire_api = "responses"

開く前に検証

curl -X POST "https://proxystation.co/v1/responses" \
  -H "Authorization: Bearer sk-XXXXXXXXXXXXXXXX" \
  -H "content-type: application/json" \
  -d '{"model":"gpt-5.6-luna","input":"ok"}'

期待される応答： A body with "status":"completed" — the Responses API is what Codex speaks.

  • wire_api accepts only "responses". Chat Completions support was removed from Codex; a config that names it will not start.
  • env_key names an environment variable, it does not hold the key. The string here must match the variable you exported, or Codex reports no credential.
  • The base URL keeps /v1 here, unlike Claude Code above. Same key, same host, different string.

aider

https://proxystation.co/v1

インストール

python -m pip install aider-install && aider-install

設定~/.aider.conf.yml

export OPENAI_API_BASE=https://proxystation.co/v1
export OPENAI_API_KEY=sk-XXXXXXXXXXXXXXXX

aider --model openai/gpt-5.6-luna

開く前に検証

curl "https://proxystation.co/v1/models" -H "Authorization: Bearer sk-XXXXXXXXXXXXXXXX"

期待される応答： A JSON list under "data" — aider reads the same catalogue.

  • OPENAI_API_BASE, not OPENAI_BASE_URL. aider passes the value to litellm, which uses the API_BASE name.
  • The model needs the openai/ prefix so litellm routes it to the OpenAI-compatible client rather than guessing the vendor from the name.

エディタと拡張機能

Continue

https://proxystation.co/v1

インストール

Install the Continue extension from the VS Code or JetBrains marketplace.

設定~/.continue/config.yaml

# ~/.continue/.env
PROXYSTATION_API_KEY=sk-XXXXXXXXXXXXXXXX

# ~/.continue/config.yaml
models:
  - name: Proxy Station
    provider: openai
    model: gpt-5.6-luna
    apiBase: https://proxystation.co/v1
    apiKey: ${{ secrets.PROXYSTATION_API_KEY }}

開く前に検証

curl -X POST "https://proxystation.co/v1/chat/completions" \
  -H "Authorization: Bearer sk-XXXXXXXXXXXXXXXX" \
  -H "content-type: application/json" \
  -d '{"model":"gpt-5.6-luna","messages":[{"role":"user","content":"ok"}],"max_tokens":8}'

期待される応答： A body with "choices" — the same call the extension makes.

  • apiBase keeps /v1 even when you set provider: anthropic. Continue appends only the method path, unlike the Anthropic SDK.
  • The same config.yaml is read by the VS Code extension, the JetBrains plugin and the cn CLI — configure once.

Cline

https://proxystation.co/v1

インストール

Install the Cline extension from the VS Code marketplace.

設定~/.cline/data/settings/providers.json

In the extension: Settings → API Provider → OpenAI Compatible

  Base URL   https://proxystation.co/v1
  API Key    sk-XXXXXXXXXXXXXXXX
  Model ID   gpt-5.6-luna

The CLI and the extension share ~/.cline/data/settings/providers.json.

開く前に検証

curl -X POST "https://proxystation.co/v1/chat/completions" \
  -H "Authorization: Bearer sk-XXXXXXXXXXXXXXXX" \
  -H "content-type: application/json" \
  -d '{"model":"gpt-5.6-luna","messages":[{"role":"user","content":"ok"}],"max_tokens":8}'

期待される応答： A body with "choices" before you touch the extension at all.

  • There is no VS Code settings.json key for this. Settings named cline.openAiCompatible.* do not exist — configure it in the extension panel.
  • Choose the "OpenAI Compatible" provider, not "OpenAI". The plain OpenAI provider ignores a custom base URL.

Cursor

https://proxystation.co/v1

インストール

Cursor itself is the editor; nothing extra to install.

設定

Settings → Models → API Keys

  Override OpenAI Base URL    https://proxystation.co/v1
  OpenAI API Key              sk-XXXXXXXXXXXXXXXX

Then press Verify.

開く前に検証

curl -X POST "https://proxystation.co/v1/chat/completions" \
  -H "Authorization: Bearer sk-XXXXXXXXXXXXXXXX" \
  -H "content-type: application/json" \
  -d '{"model":"gpt-5.6-luna","messages":[{"role":"user","content":"ok"}],"max_tokens":8}'

期待される応答： A body with "choices". Cursor keeps the key in secure storage, so this curl is the only way to check the endpoint separately from the editor.

  • No environment variable and no config file. The key lives in Cursor’s secure storage, so there is nothing to edit by hand and nothing to commit by accident.
  • Add the model id to the model list first, or the override has nothing to apply to.

SDK

OpenAI SDK

https://proxystation.co/v1

インストール

pip install openai · npm install openai

設定

export OPENAI_BASE_URL=https://proxystation.co/v1
export OPENAI_API_KEY=sk-XXXXXXXXXXXXXXXX

# or in code
from openai import OpenAI
client = OpenAI(api_key="sk-XXXXXXXXXXXXXXXX", base_url="https://proxystation.co/v1")

開く前に検証

curl "https://proxystation.co/v1/models" -H "Authorization: Bearer sk-XXXXXXXXXXXXXXXX"

期待される応答： A JSON list under "data".

  • OPENAI_BASE_URL ends in /v1. The SDK appends /chat/completions to whatever you give it.

Anthropic SDK

https://proxystation.co

インストール

pip install anthropic · npm install @anthropic-ai/sdk

設定

export ANTHROPIC_BASE_URL=https://proxystation.co
export ANTHROPIC_API_KEY=sk-XXXXXXXXXXXXXXXX

# or in code — note the base URL carries no /v1
from anthropic import Anthropic
client = Anthropic(api_key="sk-XXXXXXXXXXXXXXXX", base_url="https://proxystation.co")

開く前に検証

curl -X POST "https://proxystation.co/v1/messages" \
  -H "x-api-key: sk-XXXXXXXXXXXXXXXX" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-sonnet-5","max_tokens":8,"messages":[{"role":"user","content":"ok"}]}'

期待される応答： A body starting {"id":"msg_…

  • No /v1 here. The SDK adds it, and a base URL that already carries one produces /v1/v1/messages — a 404.

このページの設定はすべて 2026-08-30 にこのエンドポイントに対して検証済みで、上記の検証コマンドはすべて実際に実行されました。ツールは変わります。Codex は今年 Chat Completions のサポートを削除しました。合わなくなったときに教えてくれるのが検証コマンドです。

ステーションは地域別のエントリポイントでも応答します。同じキーで動作しますが、ステータスページに掲載している稼働率とレイテンシはこのオリジンに対してのみ計測しています —— それらの数値は別のエントリポイントを表しません。

API リファレンスモデルの稼働状況