In Cherry Studio’s “設定”, people commonly look for two things: provider settings to use models with their own API key, and MCP server settings to connect external tools. The former is under 設定 → モデルプロバイダー → プロバイダーを追加; the latter is under 設定 → MCP サーバー.This page explains both procedures based on v2.1.4, released on September 30, 2026, reproducing the Japanese interface labels exactly as displayed. Version 2 significantly changed the screen for adding a provider, so v1-era guides that say to select “タイプ：OpenAI” no longer match the interface.
This guide covers the desktop edition of CherryHQ/cherry-studio (AGPL-3.0; Windows, macOS, and Linux). As of October 1, 2026, the repository was not archived and the latest version was v2.1.4. The App Store app with the same name is unrelated and comes from a different developer. To switch the interface to Japanese, select Japanese in the language settings (the UI supports 13 languages, including Japanese).
Provider settings: use models with your own API key
Here is the procedure at a glance. Labels match the v2.1.4 Japanese UI.
設定 → モデルプロバイダー → プロバイダーを追加
（ダイアログ名：カスタムプロバイダーを追加）
プロバイダー名 Kunavo
APIキー sk-kn-...
エンドポイント設定
OpenAI https://api.kunavo.com/v1
Anthropic メッセージ https://api.kunavo.com
その他のオプション
OpenAI レスポンス https://api.kunavo.com/v1 （任意）
画像生成ベースURL https://api.kunavo.com/v1 （任意）
Google Gemini 空欄のまま
→ 保存 → モデル一覧で「モデルを同期」→ 使うモデルを追加 → 「チェック」- Open Settings → Model Providers and click Add Provider. The dialog that opens is titled “Add Custom Provider.” To start from an existing provider for Coding Plan services, multiple accounts, or separate projects, you can also use “Start from Preset (Optional)” at the top.
- Enter the Provider Name and API Key.
- Endpoint Settings includes two fields by default: OpenAI and Anthropic Messages. You must configure at least one text endpoint. Fill in both to make models available for Agents and features that use Anthropic format, as well as chat.
- Expand More Options to see fields for OpenAI Responses, Google Gemini, Image Generation Base URL, and Image Editing Base URL. Leave any fields you don’t need blank.
- After saving, check that the provider is enabled on the provider screen. According to the official documentation, models won’t appear in the selector if the provider is configured but disabled. This is the most common cause of a “key not working.”
- Fetch models with Sync Models in the model list, add the ones you’ll use, then test one with Check.
How to enter URLs: use only the root URL
In the v2.1.4 source, if a root URL entered in a field does not include a version segment, /v1 is added; if it already includes one, nothing is added. The path for that field is appended afterward. Each field displays the final URL as its “Request Path,” so check it before saving.
| Field | Path Cherry Studio appends | Kunavo |
|---|---|---|
| OpenAI | /chat/completions | Support |
| Anthropic Messages | /messages | Support |
| OpenAI Responses (More Options) | /responses | Support |
| Image Generation Base URL (More Options) | /images/generations | Support |
| Image Editing Base URL (More Options) | /images/edits | Support |
| Google Gemini (More Options) | /models/{model}:generateContent | Unsupported — leave blank |
Avoid two mistakes. Pasting a complete URL that includes /chat/completions or /messages duplicates the path and returns a 404. The # at the end is a symbol that, as the interface tip says, “disables automatic API version appending.” Adding it to a standard endpoint removes /v1.
Set default models to avoid unnecessary costs
Cherry Studio calls models in the background as well as for chat. As the interface description says, the Fast Model is “used for simple tasks such as naming topics and extracting search keywords.” Its hint also says, “Select a lightweight model and avoid reasoning models.” Setting an inexpensive model here helps prevent the more expensive model from running for every conversation. You can also configure a separate Translation Model. Keep in mind that selecting multiple models to answer at once sends a separate request to each model, and each request is billed separately. The usage statistics in the app show estimates based on public prices, so they will be higher than your actual cost when you use a discounted route. Update the unit prices in the model settings to match your rates. For more details, see Cherry Studio API cost in English.
MCP server settings: connect external tools
MCP is a way to connect models (Agents) to external tools and data. The official documentation gives these steps: Settings → MCP → MCP Servers → Add. You can create a server by entering only the connection details in “Quick Create,” then adjust the rest later.
| Type (as shown in the interface) | Use case | What to enter |
|---|---|---|
| Standard input/output (stdio) | A server that runs through a local command | Command, arguments, environment variables |
| Server-Sent Events (SSE) | A remote service that provides an SSE URL | URL (and authentication, if needed) |
| Streamable HTTP | A remote service that provides a Streamable HTTP URL | URL (and authentication, if needed) |
種類 標準入力/出力 (stdio)
コマンド npx
引数 -y @modelcontextprotocol/server-filesystem /Users/you/notes
環境変数 （サーバーが求めるものだけ）- Select the type that matches the connection method specified by the service. The documentation also instructs you to “follow the provider’s configuration instead of guessing from the name.”
- Save and enable the server, then wait for its status to show as healthy. Check the “Tools,” “Prompts,” and “Resources” tabs under its details to see what it provides.
- Enable the server under Work → Agent menu → Edit → MCP. Servers are not automatically attached to every Agent.
- You can also insert MCP prompts or MCP resources provided by the server using “+” in the input field.
The model calls MCP tools, so select a model that supports tool calling. The Claude and GPT models added in the provider settings above support tool calling. As the documentation recommends, enable one at a time at first and verify it works. Keep tools that can make changes or incur charges set to require approval. If you install tools from the built-in MCP server or marketplace, check the commands and environment variables too.
Using Kunavo: notes and payment
- Verification scope. This setup is based on Cherry Studio’s source code and official documentation; Kunavo has not tested Cherry Studio connected to its own endpoints. Keep your current working route and try this one alongside it.
- Kunavo supports chat and images only. There is no embedding model, so vector search in a knowledge base requires another provider or a local embedding model (the documentation says keyword search with BM25 still works without embeddings).
- Payment. Prepaid top-ups with no monthly fee; the balance is deducted per token. The minimum top-up is $10, and Stripe Checkout supports cards (Visa, Mastercard, American Express, JCB), Apple Pay, Google Pay, and Link. See billing details, then create an account and issue a key. The English setup page is Cherry Studio integration guide.
Frequently asked questions
How do I set up my own API key in Cherry Studio?
Open the “Add Custom Provider” dialog at Settings → Model Providers → Add Provider. Enter a provider name and API key, then enter a root URL in the OpenAI and Anthropic Messages fields under Endpoint Settings and save. Next, fetch models using “Sync Models” in the model list, add the ones you’ll use, and test one with “Check.” Note that the provider must be enabled for its models to appear in the model selector.
Does Cherry Studio’s API address need /v1?
Either works. In the v2.1.4 source, if the root URL you enter does not include a version segment (/v1), it is added automatically; if it is already there, it is left as is. The path for each field is then appended (for example, /chat/completions for OpenAI and /messages for Anthropic). Avoid pasting a complete URL that already includes /chat/completions, as the path will be duplicated and return a 404. The # at the end disables automatic version appending, so don’t add it to standard endpoints. You can check the final URL under “Request Path” beneath each field.
Where do I configure an MCP server in Cherry Studio?
The official documentation’s steps are Settings → MCP → MCP Servers → Add. Local commands typically use standard input/output (stdio); remote services typically use SSE or Streamable HTTP. Enter the details as provided by the service. After saving, enable the server and check its available tools in the “Tools” tab under its details. Then enable the server under Work → Agent menu → Edit → MCP. The model calls the tools, so select one that supports tool calling.
Is Cherry Studio free?
The desktop (Community) edition is free, open-source software under AGPL-3.0. You pay for model usage through the provider you configure. Cherry Studio Enterprise is a separate commercial product with custom pricing; CherryAI, which is built in, is free, but its model lineup and limits are not public.
What if “Sync Models” returns nothing?
This button uses the address and key you entered to fetch the provider’s model list (/v1/models). If nothing appears, first check the address or key. Make sure you haven’t pasted a complete URL or added # at the end, and try the same combination with curl. If it returns JSON, the issue is on the app side; if it returns 401, the key is the problem.
Verified on October 1, 2026: GitHub API (CherryHQ/cherry-studio, v2.1.4), v2.1.4 Japanese UI strings (ja-jp.json) and source code for the Add Provider screen, and Cherry Studio’s official MCP documentation. Kunavo has not run Cherry Studio against its own endpoints.