> ## Documentation Index
> Fetch the complete documentation index at: https://anam.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP server

> Connect Claude, ChatGPT, Cursor, VS Code, Claude Code, Codex or any MCP client to your Anam account, and build and manage personas from chat.

Anam's MCP server lets your AI app build and manage personas (real-time AI avatars that talk with people over video), review their conversations and write code to add them to your app. Anam hosts it; there's nothing to install.

```text Server URL theme={"system"}
https://lab.anam.ai/mcp
```

You sign in with your Anam account and choose one organization. The connection reaches only that organization.

## Connect an app

The first time, your app opens Anam in a browser. Sign in or sign up, choose the organization and choose the access to give it (see [Access and safety](#access-and-safety)).

> [tab] Claude
>
> These steps work on claude.ai and in the Claude desktop app.
>
> **1. Add a custom connector**
>
> In Claude, open **Settings**, then **Connectors**, and choose **Add custom connector**.
>
> **2. Paste the server URL**
>
> Name it Anam, paste `https://lab.anam.ai/mcp` and add it.
>
> **3. Sign in**
>
> Choose **Connect** and sign in to Anam.
>
> On Team and Enterprise plans, an owner adds the connector in the organization's settings first.

> [tab] ChatGPT
>
> **1. Add a custom MCP server**
>
> In ChatGPT, open **Plugins**, choose the plus button, then **Add custom MCP server**.
>
> **2. Paste the server URL**
>
> Name it Anam, paste `https://lab.anam.ai/mcp` and choose **OAuth** for authentication.
>
> **3. Sign in**
>
> Accept the warning, choose **Create as a plugin** and sign in to Anam.
>
> **4. Use it in a chat**
>
> Install the plugin, start a new conversation, type `@` and choose Anam.
>
> Workspace admins can restrict custom MCP servers.

> [tab] Claude Code
>
> Run these in a terminal to add Anam for all your projects and sign in:
>
> ```bash theme={"system"}
> claude mcp add --transport http --scope user anam https://lab.anam.ai/mcp
> claude mcp login anam
> ```
>
> You can also run `/mcp` in Claude Code and choose anam to sign in.

> [tab] Cursor
>
> [Add to Cursor](cursor://anysphere.cursor-deeplink/mcp/install?name=anam\&config=eyJ1cmwiOiJodHRwczovL2xhYi5hbmFtLmFpL21jcCJ9)
>
> Or add Anam to `~/.cursor/mcp.json`, or to `.cursor/mcp.json` in a project:
>
> ```json mcp.json theme={"system"}
> {
>   "mcpServers": {
>     "anam": {
>       "url": "https://lab.anam.ai/mcp"
>     }
>   }
> }
> ```
>
> Cursor asks you to sign in to Anam when it first connects.

> [tab] VS Code
>
> [Add to VS Code](vscode:mcp/install?%7B%22name%22%3A%22anam%22%2C%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Flab.anam.ai%2Fmcp%22%7D)
>
> Or run:
>
> ```bash theme={"system"}
> code --add-mcp '{"name":"anam","type":"http","url":"https://lab.anam.ai/mcp"}'
> ```
>
> Or add Anam to `.vscode/mcp.json` in your workspace and start the server from the file:
>
> ```json .vscode/mcp.json theme={"system"}
> {
>   "servers": {
>     "anam": {
>       "type": "http",
>       "url": "https://lab.anam.ai/mcp"
>     }
>   }
> }
> ```
>
> VS Code opens Anam so you can sign in.

> [tab] Codex
>
> Run these in a terminal to add Anam to Codex and sign in:
>
> ```bash theme={"system"}
> codex mcp add anam --url https://lab.anam.ai/mcp
> codex mcp login anam
> ```

> [tab] Other clients
>
> Add `https://lab.anam.ai/mcp` as a remote MCP server (Streamable HTTP).
>
> Sign-in is OAuth 2.1 with PKCE. The server's 401 points to its protected resource metadata at `https://lab.anam.ai/.well-known/oauth-protected-resource/mcp`, and clients register themselves, by dynamic client registration or a client ID metadata document. The scopes are `anam:read`, `anam:write` and `offline_access`.
>
> Platforms that need a client ID and secret made in advance, such as Salesforce Agentforce, AWS AgentCore Gateway, Microsoft 365 Copilot or Google, use an agent client: owners and admins create one on the [AI apps page](https://lab.anam.ai/ai-apps) in Anam Lab.

To check the connection, ask your app "Which Anam organization am I connected to?" It runs `anam_connection_info`, which returns the organization, the access you chose and the app. If it can't connect, see [Troubleshooting](#troubleshooting).

## Agents without a browser: API keys

Agents that run without a browser, such as CI jobs, can send an Anam API key as `Authorization: Bearer <key>` instead of signing in. They act with the key's permissions and don't appear under Connected apps on the AI apps page. Create and revoke keys on the [API keys page](https://lab.anam.ai/api-keys).

Keep the key in an `ANAM_API_KEY` environment variable and have the client read it from there, so it never sits in a config file:

```json Cursor (mcp.json) theme={"system"}
{
  "mcpServers": {
    "anam": {
      "url": "https://lab.anam.ai/mcp",
      "headers": {
        "Authorization": "Bearer ${env:ANAM_API_KEY}"
      }
    }
  }
}
```

```json Claude Code (.mcp.json) theme={"system"}
{
  "mcpServers": {
    "anam": {
      "type": "http",
      "url": "https://lab.anam.ai/mcp",
      "headers": {
        "Authorization": "Bearer ${ANAM_API_KEY}"
      }
    }
  }
}
```

```bash Codex theme={"system"}
codex mcp add anam --url https://lab.anam.ai/mcp --bearer-token-env-var ANAM_API_KEY
```

## Access and safety

When you connect, you choose one organization and, if the app asks for both, its access:

| Access | What the app can do |
| - | - |
| Full access | Create, change and delete things and start work billed to the organization, as a full-access API key can. |
| Read only | Change nothing, but read everything in the organization, including session transcripts, recordings and knowledge. |

An app Anam can't verify is marked Unverified on the sign-in page and starts at Read only; choose Full access only if you trust it.

* **Confirmations.** Before the app deletes something, starts paid work (creating an avatar or avatar video, cloning a voice, sending a persona to a meeting), generates speech (importing a voice or changing its model) or adds a knowledge document (documents count towards your monthly upload allowance), Anam asks you to confirm in the app, and the action runs only if you approve. Apps that can't show Anam's question (MCP elicitation) are instead told to ask you first; Anam can't enforce that, so it depends on the app.
* **Same rules as the API.** Every tool runs Anam's public API with your connection's own key, so the same permissions, organization boundaries, plan limits, rate limits and audit log apply.
* **Secrets stay hidden.** Results never show stored API keys, and webhook headers, tokens and similar values show as `••••`. LLM provider keys you add are stored encrypted and never returned.
* **Untrusted content.** The app is told to treat what tools return (transcripts, documents, web pages) as data, not instructions.
* **Docs searches.** `anam_docs_search` and `anam_docs_read` send the search text or read command to Anam's public docs, hosted by Mintlify, with no account details. Send documentation terms, not customer content or secrets.

## Tools

* **Personas:** create one from a short brief or from Anam's defaults, change its prompt, avatar, voice and language model, and get a share link to talk to it.
* **Avatars, voices and LLMs:** browse what your organization can use, create an avatar from an image, import a voice from a provider or clone one from a recording, and add your own LLM.
* **Knowledge:** make folders, add documents from text or web pages, search them and attach them to personas.
* **Sessions:** find conversations by date, persona or what was said, read transcripts, and see reply speed and failures.
* **Meetings and video:** send a persona to a Google Meet, Zoom or Microsoft Teams call, and generate avatar videos.
* **Building:** starter code for Next.js, Express or FastAPI, integration guides and search over these docs.

Conversations with a persona happen in a browser or your own app, not in the chat. Your Anam plan's limits and charges apply, as in Anam Lab.

Most tools run one Anam API operation; the workflow tools under Start here do several steps at once. A read-only connection gets only the read tools. Apps installed from a directory listing get a smaller set made for that platform (see Directory tool sets below); everywhere else, use `/mcp`, which has every tool.

> [accordion] Start here
>
> * `anam_connection_info`: the connected organization, access and app.
> * `anam_usage_get`: plan, minutes used and left, live sessions against the limit, and knowledge uploads this month.
> * `anam_options`: avatars, voices and LLMs in one list, with Anam's recommended picks.
> * `anam_quickstart`: a working persona and a share link in one call, from Anam's defaults or a template (sales, support, receptionist, tutor, interviewer).
> * `anam_persona_from_brief`: a persona with a written system prompt and greeting from a description.

> [accordion] Personas and share links
>
> * `anam_persona_list`, `anam_persona_get`, `anam_persona_create`, `anam_persona_update`, `anam_persona_delete`
> * `anam_persona_widget_get`: a persona's widget settings.
> * `anam_share_link_list`, `anam_share_link_get`, `anam_share_link_create`, `anam_share_link_update`, `anam_share_link_delete`: public links anyone can open to talk to a persona.

> [accordion] Avatars, voices and LLMs
>
> * `anam_avatar_list`, `anam_avatar_get`, `anam_avatar_create` (from an image URL), `anam_avatar_update`, `anam_avatar_delete`
> * `anam_avatar_video_create`, `anam_avatar_video_list`, `anam_avatar_video_get`, `anam_avatar_video_cancel`, `anam_avatar_video_retry`, `anam_avatar_video_delete`
> * `anam_voice_list`, `anam_voice_get`, `anam_voice_import`, `anam_voice_update`, `anam_voice_delete`
> * `anam_voice_clone`: a new voice cloned from a recording at a public URL (10 to 15 seconds of one person speaking, up to 10 MB). Only clone the voice of someone who has consented.
> * `anam_llm_list`, `anam_llm_get`, `anam_llm_create`, `anam_llm_update`, `anam_llm_delete`: your own LLM endpoints.
> * `anam_pronunciation_dictionary_list`, `anam_pronunciation_dictionary_get`, `anam_pronunciation_dictionary_create`, `anam_pronunciation_dictionary_import`, `anam_pronunciation_dictionary_update`, `anam_pronunciation_dictionary_delete`

> [accordion] Knowledge and persona tools
>
> * `anam_knowledge_folder_list`, `anam_knowledge_folder_get`, `anam_knowledge_folder_create`, `anam_knowledge_folder_update`, `anam_knowledge_folder_delete`, `anam_knowledge_folder_search`
> * `anam_knowledge_document_create_from_text`, `anam_knowledge_document_create_from_url`: add a document; it becomes searchable once processed.
> * `anam_knowledge_document_list`, `anam_knowledge_document_get`, `anam_knowledge_document_update`, `anam_knowledge_document_delete`, `anam_knowledge_document_download`
> * `anam_persona_knowledge_add`, `anam_persona_knowledge_remove`: attach or detach knowledge folders, as the Build page does.
> * `anam_tool_list`, `anam_tool_get`, `anam_tool_create`, `anam_tool_update`, `anam_tool_delete`: webhooks and client tools a persona can call.

> [accordion] Sessions and meetings
>
> * `anam_session_search`: find sessions by date, persona or what was said, or page through one transcript.
> * `anam_session_insights`: how one call, or every call in a period, went: reply speed, interruptions, errors and failed tool calls.
> * `anam_session_list`, `anam_session_get`, `anam_session_transcript_get`, `anam_session_analytics_get`, `anam_session_stats`, `anam_session_recording_get`, `anam_session_concurrency_get`
> * `anam_session_stop`: end a live session. `anam_session_delete`: delete a session, its recording or its report.
> * `anam_session_token_create`: a short-lived session token for trying an app locally. It starts at most 10 sessions unless you ask for more (up to 100).
> * `anam_meeting_invite_create`, `anam_meeting_invite_list`, `anam_meeting_invite_get`, `anam_meeting_invite_cancel`: send a persona to a Google Meet, Zoom or Microsoft Teams meeting.

> [accordion] Building and docs
>
> * `anam_integration_guide`: the session-token pattern and a docs excerpt for JavaScript, React, LiveKit, Python, ElevenLabs or the widget.
> * `anam_scaffold`: starter files for Next.js, Express or FastAPI, with your persona's ID filled in and the API key kept on the server.
> * `anam_docs_search`, `anam_docs_read`: search and read these docs.

> [accordion] Directory tool sets
>
> Each directory listing connects to its own path, with the tools that platform's rules allow. A token issued for one path doesn't work on another.
>
> * `/mcp/claude`: the Claude connectors directory
> * `/mcp/chatgpt`: the ChatGPT plugin directory
> * `/mcp/copilot`: Microsoft 365 Copilot
> * `/mcp/gemini`: Google's Gemini
> * `/mcp/agentforce`: Salesforce Agentforce
> * `/mcp/muse`: Meta Muse

## Disconnect or turn off AI apps

* [AI apps](https://lab.anam.ai/ai-apps) in Anam Lab lists the organization's connected apps, with their access, who connected them and when they were last used. Disconnecting stops an app at once; it has to sign in again to come back. Members disconnect their own apps, and owners and admins anyone's.
* Owners and admins can turn off **Allow AI apps** on the same page. That disconnects every member's apps and the organization's agent clients, and refuses MCP requests with its API keys (the keys keep working for the rest of the API). Nobody can connect an app until it's turned on again. Organizations that require SSO start with AI apps off.
* Resetting your password or signing out everywhere also disconnects your apps. For an agent that uses an API key, revoke the key.

## Docs-only MCP server

To let an assistant search and read these docs without connecting an Anam account, add the docs MCP server instead. It needs no sign-in and can't reach your account.

```text Docs server URL theme={"system"}
https://anam.ai/docs/mcp
```

```bash Claude Code theme={"system"}
claude mcp add --transport http anam-docs https://anam.ai/docs/mcp
```

The account server above already includes docs search (`anam_docs_search` and `anam_docs_read`), so you don't need both.

## Troubleshooting

| Problem | What to do |
| - | - |
| You can't choose your organization when signing in | AI apps are off for it. An owner or admin turns on **Allow AI apps** on the [AI apps page](https://lab.anam.ai/ai-apps). Organizations that require SSO start with them off. |
| The app asks you to connect again | Its connection ended, for example because someone disconnected it, AI apps were turned off, or you reset your password or signed out everywhere. Sign in again. |
| A tool you need is missing | A Read only connection gets only read tools: disconnect the app on the [AI apps page](https://lab.anam.ai/ai-apps), connect again and choose Full access. Apps installed from a directory listing get a smaller set; for every tool, add Anam as a custom connector at `https://lab.anam.ai/mcp`. |
| The app works in the wrong organization | Disconnect it on the [AI apps page](https://lab.anam.ai/ai-apps), connect again and choose the right organization. |
| Requests with an API key get `401` | The key is wrong or revoked, or `ANAM_API_KEY` isn't set in the environment the client starts from. Check the variable and restart the client, or create a new key. |
| Requests with an API key get `403` | AI apps are off for the key's organization, so MCP refuses its keys (turn on **Allow AI apps**), or the key works only with the ElevenLabs adapter. The error says which. |
| Confirmed actions never run in `claude -p` | Non-interactive Claude Code dismisses Anam's confirmation questions, so deletes and paid work don't run. Run the task interactively, or add an `Elicitation` hook that answers them. |
| A client can't connect to `https://docs.anam.ai/mcp` | That old docs server address redirects, and MCP clients don't follow the redirect. Use `https://anam.ai/docs/mcp`. |
| You still use the `anam-mcp` Python package | Remove it from your client and connect to `https://lab.anam.ai/mcp`. The hosted server replaces it, signs you in with OAuth and gets new tools without an update. |

## Help

* Support: [support@anam.ai](mailto:support@anam.ai)
* Security and how to report a vulnerability: [anam.ai/security](https://anam.ai/security)
* [Privacy policy](https://anam.ai/privacy-policy) and [terms of service](https://anam.ai/terms-of-service)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.