> ## Documentation Index > Fetch the complete documentation index at: https://docs.voiceinfra.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Use VoiceInfra from Claude, ChatGPT, and Other MCP Clients > VoiceInfra runs a remote MCP server, so you can manage agents, numbers, and call logs from any tool that speaks Model Context Protocol. Authentication is OAuth — no API key required. VoiceInfra exposes a **remote MCP server**. That means you can work with your VoiceInfra workspace from inside Claude, ChatGPT, or any other tool that supports the Model Context Protocol — asking *"list my voice agents"* or *"why did the last call end early?"* without leaving that tool. There is no API key to create and no token to paste. Your client registers itself and you approve access in the browser. > [!NOTE] > > Authentication uses OAuth. You will be asked to sign in and choose a workspace the first time you connect, and you can [disconnect at any time](https://docs.voiceinfra.ai/docs/api/connected-apps.md). ## Your MCP address Every workspace has one address to copy, shown in the portal under **Connected apps → Your MCP address**. It looks like: ```text theme={null} https://.voiceinfra.ai/mcp ``` The address is **specific to the region your account lives in**, because each region is a separate deployment with its own data. Copy it from the portal rather than constructing it by hand. ## Connecting **Claude** **1. Add a custom connector** Open Claude → **Settings → Connectors → Add custom connector**. **2. Paste your address** Paste your MCP address as the URL. **Leave the client ID and secret empty** — Claude registers itself automatically. **3. Approve access** Click **Connect**. A window opens asking you to sign in, choose a workspace, and approve the permissions. **4. Try it** Claude will list the tools it can now use. Try something like *"list my voice agents"*. > [!TIP] > > If you are using the desktop app you can close the browser window once approval is complete — the connector picks the session up. **Claude Code** Run this in your terminal, replacing the address with your own: ```bash theme={null} claude mcp add --transport http voiceinfra https://.voiceinfra.ai/mcp ``` It opens a browser to approve access. Afterwards, running `/mcp` inside Claude Code shows the connection and the tools available on it. **ChatGPT** **1. Enable developer mode** Open ChatGPT → **Settings → Connectors → Advanced → Developer mode**, then choose **Create**. **2. Add the server** Paste your MCP address as the server URL and choose **OAuth** as the authentication method. **3. Approve and enable** Approve access when prompted, then enable the connector inside a chat. **Any other MCP client** Add a **remote MCP server** using your address, and choose **OAuth** as the authentication method. No key or token is needed. Your browser opens the VoiceInfra approval screen, where you pick a workspace and allow access. ## Approving access Whichever client you use, the same VoiceInfra screen handles the connection: 1. **Sign in** if you are not already signed in. 2. **Choose a workspace** — the app acts only within the workspace you pick. 3. **Review the permissions** the app asked for, and allow or deny. Each permission is grouped into an area you will recognise, and shows whether it is **view** or **can change**. Anything that spends credit is called out separately. You can grant change access to some areas and leave others read-only. The app's ability to change anything is still limited by **your role in that workspace** — see [Connected apps](https://docs.voiceinfra.ai/docs/api/connected-apps.md). > [!WARNING] > > If the consent screen shows **"This app hasn't been reviewed by VoiceInfra"**, the app registered itself and anyone can do that. Its name and logo prove nothing. Only continue if you started the connection from the app yourself and trust it. ## What you can do from a connected client Once connected, the client can work with the same resources you can, within the workspace you chose: | Area | Examples | | - | - | | **Agents** | List agents, read their configuration, create and update them | | **Workflows** | Read and edit an agent's workflow graph | | **Phone numbers** | List numbers and show which agent answers each one | | **Knowledge** | List knowledge bases and their documents | | **Functions & MCP servers** | Inspect custom functions and connected MCP servers | | **Call logs** | Read a call, its transcript, summary, and cost | | **Analytics** | Report on call volume and agent activity | | **Billing** | Read your subscription and current usage | | **Versions** | Publish, roll back, and discard drafts | Actions that change your configuration or spend money respect the same approval rules as elsewhere in the product, and are governed by the permissions you granted. ## Disconnecting Open **Connected apps** in the portal to see every client you have connected, when it was added, and what it can do. **Disconnect** takes effect immediately — the client stops working right away and anyone using it must connect again. ## Frequently asked questions > **Do I need an API key?** > > No. The MCP server uses OAuth. Your client registers itself and you approve access in the browser. For programmatic access with a key instead, see the [API reference](https://docs.voiceinfra.ai/docs/api/overview.md). > **Why is my address tied to a region?** > > Each region is a separate deployment holding its own data. Your MCP address points at the region your account lives in, which is why you should copy it from the portal rather than guess it. > **Can the client see other workspaces?** > > Only the workspace you chose on the approval screen. To work with another workspace, connect again and pick it. > **Can an AI client make changes without asking me?** > > It can only use the permissions you granted. Change access is opt-in per area on the consent screen, and spending credit is always surfaced separately. What it can actually change is also bounded by your workspace role. > **Does this cost anything extra?** > > No. Using the MCP server is included. Calls and messages you place through it are metered normally. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.