# Connect Turguz to your AI assistant > Manage your servers, projects and databases from the assistant you use. Setup steps for each assistant, the values for assistants not listed, and connecting with an access key. Turguz supports the open standard that AI assistants use to connect to external tools (MCP, Model Context Protocol). Once connected, you do the work you do in the panel by telling your assistant: creating a server, deploying a project, setting up a database, checking your quota and usage. The assistant signs in to Turguz with your account. Every request goes through the panel's permission checks; anything you cannot do in the panel, you cannot do with the assistant either. ## What you can do You can tell your assistant things like: - "Which projects do I have on Turguz, and which ones are live?" - "Deploy the latest version to my project's dev environment, and tell me the address once the site is up." - "Create a small server with Ubuntu installed and give me the command to connect to it." - "Connect a database to this environment." - "Why did the last deployment fail? Check the log." The tools the assistant can use match the sections of the panel: account, servers and kept disks, leases and quota, projects, environments and deployments, databases, internet access, published services and domains, allowed addresses, team members and notifications. The platform admin also has administration tools. Some steps the assistant cannot finish for you: you approve the Git page that opens to grant access to your repository, and you write the DNS record for your own domain. The assistant tells you what to do. For a root domain (e.g. `example.com`) the assistant first suggests the www name; if you connect the root domain directly without the ownership record, the connection waits for an administrator's approval. You can also upload your project with the assistant. If the assistant can run commands on your computer (for example, assistants in a terminal), it uploads the project folder in one go, with the same limit as the panel's «Arşiv yükle» (Upload archive): it gets an upload link that is valid for a short time and works once, and sends the archive straight from your computer to Turguz. The file contents do not go to the assistant for the upload. An assistant that cannot run commands can only upload small projects it wrote itself; for larger ones, run the command the assistant gives you yourself, or upload the project from the panel. ## Before you start - A Turguz account and a role in at least one tenant. «Kiracı yöneticisi» (Tenant admin) and «Geliştirici» (Developer) can make changes; «İzleyici» (Viewer) can only view. - Turguz's MCP address: `https://mcp.turguz.app/mcp` ## How to add it Each assistant has its own section below: depending on what the assistant supports, a one-click link, a name and address to enter in the app's own add-server screen, a command to run in a terminal, or a block to add to a settings file. Only setup routes actually tested with Turguz are listed here; new ones are added as they are tested. If your assistant is not listed, use the values in the **Other assistants** section; an assistant that cannot sign in through a browser connects [with an access key](#erisim-anahtariyla-baglanmak). ### Claude Code **Add Turguz** Run in a terminal: ```bash claude mcp add turguz https://mcp.turguz.app/mcp --transport http --scope user ``` **Connect** Open Claude Code, type `/mcp`, select **turguz** from the list and press **Authenticate**. Turguz's sign-in page opens in your browser; sign in with the account you use in the panel. If the browser does not open by itself, open the address Claude Code prints on screen in your browser. If you prefer to sign in from the terminal: `claude mcp login turguz` ### Claude desktop In the **Code** tab you add Turguz from a terminal: run the command in the Claude Code section once in your computer's terminal. Then type `/mcp` in the Code tab, select **turguz** from the list and start the sign-in. Turguz's sign-in page opens in your browser; sign in with the account you use in the panel. ### VS Code **Add Turguz** Add an HTTP server with VS Code's command for adding an MCP server; enter `Turguz` as the name and `https://mcp.turguz.app/mcp` as the address. **Connect** You are asked to sign in when you first start the server. Turguz's sign-in page opens in your browser; sign in with the account you use in the panel. ### Codex **Add Turguz** Codex's desktop app comes as **ChatGPT**: in its settings, open the MCP servers section, add a new MCP server of the streamable HTTP type, enter `Turguz` as the name and `https://mcp.turguz.app/mcp` as the address, then start authentication. **Connect** Turguz's sign-in page opens in your browser; sign in with the account you use in the panel. ### Antigravity **Open the configuration** File path: - All projects: `~/.gemini/antigravity/mcp_config.json` **Add Turguz** ```json { "mcpServers": { "turguz": { "serverUrl": "https://mcp.turguz.app/mcp" } } } ``` **Connect** If the file already has `mcpServers`, add only the `turguz` entry inside it. In Antigravity, find **turguz** in the **MCP Servers** list and press **Authenticate**. Turguz's sign-in page opens in your browser; sign in with the account you use in the panel. After you approve, copy the code shown in the browser and paste it into the **Enter authorization code** field in Antigravity. Antigravity stores your sign-in in `~/.gemini/antigravity/mcp_oauth_tokens.json` with permissions that let anyone on the computer read it. After signing in, run this once in a terminal: `chmod 600 ~/.gemini/antigravity/mcp_oauth_tokens.json` ### Other assistants - Server address: `https://mcp.turguz.app/mcp` - Client ID: `turguz-mcp` If your assistant is not listed here, enter Turguz's address in its MCP settings as an HTTP connection. If the assistant signs in through a browser, enter `turguz-mcp` as the client ID. The return address must be on your computer: `http://localhost:/callback` or `http://127.0.0.1:/callback`. An assistant that does not let you enter a client ID cannot sign in through a browser. An assistant that cannot sign in through a browser connects with a personal access key: add the key as an `Authorization: Bearer ` header. The same instructions are also in the panel: choose your assistant in the «Asistan bağlantısı» (Assistant connection) section of the **Ayarlar** (Settings) page; the address and the command come filled in for your installation. ## Connecting with an access key **Depends on your installation:** If this feature is not enabled in your installation, it does not appear in the panel; ask your platform admin. Access keys are an installation setting; if they are turned off, the key section in Settings says so. An assistant that cannot sign in through a browser can connect to Turguz with a personal access key. You create the key in the **MCP erişim anahtarları** (MCP access keys) section of the **Ayarlar** (Settings) page: [Assistant access keys](/docs/en/hesap-ve-ekip#asistan-erisim-anahtarlari). If your assistant's section has a **Connect with an access key** step, use it. Otherwise enter Turguz's address (`https://mcp.turguz.app/mcp`) in the assistant's MCP settings and add the key as an `Authorization: Bearer …` header. For this, your assistant must allow adding custom headers to an MCP server. Don't write the key into a shared file or a file that goes into a repository; keep it in an environment variable on your computer (for example `TURGUZ_MCP_ANAHTARI`). Administrator tools are not available with a key; if you lose the key, revoke it and create a new one. ## claude.ai and the Claude apps **Depends on your installation:** If this feature is not enabled in your installation, it does not appear in the panel; ask your platform admin. The cloud assistant connection is an installation setting; if it is off, the assistant section in Ayarlar (Settings) says so. In the Claude desktop app and on the claude.ai website you can add Turguz as a custom connector. On this path the connection is made by Anthropic's servers, so you have to consent first. 1. Open the **Bulut asistanı kullanıyorum** (I use a cloud assistant) card in the **Asistan bağlantısı** (Assistant connection) section of the **Ayarlar** (Settings) page and accept the consent text. The consent is personal: it opens only your own connection, and when you gave it is recorded. 2. In Claude, go to **Settings › Connectors** and press **Add custom connector**. 3. Enter `Turguz` as the name and `https://mcp.turguz.app/mcp` as the address, then press **Connect**. Turguz's sign-in page opens in the browser; sign in with the account you use in the panel. What you need to know: - The data the assistant accesses goes to Anthropic and may be processed abroad. This connection is your decision and responsibility; you give the consent on behalf of your organisation and for your own use (the Disclaimer, clauses 8.3, 2.1 and 2.2). - Platform admins can also reach your tenant with a cloud assistant unless the tenant has banned it. - Your organisation can ban cloud assistants in a tenant. You cannot reach a banned tenant with a cloud assistant, and that tenant appears in none of the answers the assistant sees. The tenant admin sets and lifts the ban in the same section; a ban set by the platform admin can only be lifted by the platform admin. - Your sign-in is stored at Anthropic. Removing the connector in Claude's settings cuts the connection. - On this path too the assistant does what you can do in the panel, with the same permissions: it opens requests, reads your notifications and manages members and teams. Application output (logs, console, command output) and files also reach the assistant; they may contain personal data of your application's visitors. - Administration tools and tasks that need a recent sign-in are not available on this path; do them in the panel or with the assistant on your computer. - Passwords, secret values and people's first names, last names and email addresses never reach the assistant; members appear with their username, role and team. - When the scope widens, your earlier consent no longer counts; the assistant asks you to consent first, and you read the new text and consent again. - If you withdraw your consent or your tenant is banned, access is cut immediately and the assistant tells you why. ## How long sign-in lasts The assistant's sign-in expires if it is not used at all for 7 days; even if you use it, you are asked to sign in again at least once every 30 days. To stop using the assistant, remove the Turguz entry from its settings (for example in Claude Code: `claude mcp remove turguz -s user`). The sign-in itself ends on its own after 7 days unused. To cut the connection from the panel, close the assistant's session with «Kapat» (Close) in the session list on the **Ayarlar** (Settings) page: the assistant has to sign in again, and you can connect the same assistant again right away. If you think the sign-in was exposed, close the session and tell the platform admin. If you connected with an access key, revoke the key; the connection is cut off within a minute. ## Permissions and security - The assistant works with your account and does what your role allows. Administration tools work only for the platform admin; for anyone else they answer "yetkiniz yok" (you do not have permission). - Irreversible actions, such as deleting a tenant, require a recent sign-in. If the assistant tells you so, sign in again from the assistant's MCP settings (for example in Claude Code: `/mcp` → turguz → sign in again). An assistant connected with an access key cannot do these; do them in the panel. - Your applications' logs, command output and messages from external systems are passed to the assistant between markers, with a warning that they are data, not instructions. This lowers the risk but cannot rule it out: review what the assistant proposes before you approve an irreversible action. - The data your connected assistant can access is transferred to that assistant's service provider; the provider may be abroad. Connecting is your decision and your responsibility (the Disclaimer you accepted, clause 8.3). - The assistant does not work with an account that has not accepted the «Sorumluluk Reddi Beyanı» (Disclaimer), and it cannot accept the Disclaimer on your behalf. If the assistant says acceptance is needed, sign in to the panel and accept the Disclaimer. - Access records showing who connected from which address and when are never given to the assistant; they are visible only in the panel, to the platform admin. ## Troubleshooting ### The sign-in page shows a "redirect_uri" error The assistant's return address is not in the form Turguz expects. Copy the command or block again from its section on this page (or from Settings). For an assistant that is not listed, check that the return address is on your computer and ends in `/callback`. ### Tools appear twice Turguz is registered in two places under the same name. In Claude Code, check with `claude mcp list` and remove the extra one: `claude mcp remove turguz -s local` or `claude mcp remove turguz -s project`. ### Sign-in does not complete on a remote machine If Claude Code is running on another machine, open the sign-in link in your own computer's browser. After sign-in, the browser goes to an address starting with `http://localhost` and says "unable to connect"; this is expected. Copy the whole address from the address bar and paste it into Claude Code. If sign-in does not complete while you are connected to a remote machine from the Claude desktop app, open a terminal on that machine, type `claude` and sign in there with `/mcp`. ### The assistant asks you to sign in again Your sign-in has expired. Start the sign-in again from the assistant's MCP settings (for example in Claude Code with `/mcp`). ### A long wait is cut off If the assistant says a tool call was cut off because it "sent no response or progress", ask for the same step again. The work on Turguz continues; waiting tools pick up where they left off. ### The "yetkiniz yok" (you do not have permission) response That action is outside your role. Administration tools work only for the platform admin; making changes in a tenant requires the «Kiracı yöneticisi» (Tenant admin) or «Geliştirici» (Developer) role.