# Connect an AI agent > Use ChatGPT, Claude, Codex, or another agent to manage training and explore permitted workspace data. Source: https://lira.thoughtly.com/docs/workspace/ai-agents Audience: admin Last reviewed: 2026-10-22 Connect your preferred AI agent to Lira to create scenarios, manage assignments, and explore submitted results. The connection uses MCP, a standard that lets an agent work with Lira's tools on your behalf. You need an **Admin** role in the Lira workspace. Each person signs in with their own account, and Lira checks their current access on every request. ## Get your connection details Open [**Settings → Integrations → AI agents**](https://lira.thoughtly.com/settings/integrations#ai-agents) in Lira. Choose your client, then use **Copy setup** for its connection steps. The server address for this Lira site is: ```text https://lira.thoughtly.com/mcp ``` If you use more than one Lira environment, copy the address from the workspace you intend to manage. You do not need to create or paste an API key. ## Connect your agent Choose the app you use. Keep Lira open while you complete the sign-in in your browser. **Codex** Run these commands in your terminal, then sign in with your Lira admin account. ```sh codex mcp add lira --url 'https://lira.thoughtly.com/mcp' codex mcp login lira ``` 1. In Codex, ask Lira to show your workspace and available tools. If Lira is already configured, skip the add command and run `codex mcp login lira`. You can also manage the connection in Codex's MCP settings. [Codex MCP documentation](https://developers.openai.com/codex/mcp/) **Claude** **Claude Code** Add Lira to Claude Code from your terminal. ```sh claude mcp add --transport http --scope user lira 'https://lira.thoughtly.com/mcp' ``` 1. Open Claude Code, run /mcp, and choose Lira to sign in with your admin account. 2. Ask Lira to show your workspace and available tools. If Lira is already listed, use `/mcp` to authenticate it instead of adding it again. Trust only a project folder you recognize when Claude asks for workspace access. [Claude Code MCP documentation](https://code.claude.com/docs/en/mcp) **Claude web, Desktop, and Cowork** Open **Customize → Connectors**, choose **Add custom connector**, and name it **Lira**. Enter the server address above, keep OAuth authentication, and complete sign-in. For Team or Enterprise workspaces, an authorized owner may need to add the connector in **Organization settings → Connectors** first. Enable Lira in the conversation's connectors menu before using it. These remote connectors require a server reachable from Anthropic's cloud, including when you use the Desktop app. [Claude custom connector documentation](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp) **ChatGPT** Add Lira as a custom MCP server in ChatGPT. Availability depends on your ChatGPT plan and workspace permissions. 1. In ChatGPT, open Plugins, select +, then Add custom MCP server. 2. Name it Lira, set the server URL to https://lira.thoughtly.com/mcp, and select OAuth authentication. 3. Review the connection and tools, then sign in with your Lira admin account. 4. If you use the legacy Apps settings, choose Apps → Create, then scan the tools and sign in. Custom MCP availability depends on your ChatGPT plan and workspace permissions. An authorized workspace admin may need to create the connection. Review the requested tools and the per-server risk notice before confirming. ChatGPT needs a server reachable from its cloud. A private development address or VPN-only address may work in a local coding client while remaining unavailable to ChatGPT. Ask your Lira administrator for the public server address if this happens. [ChatGPT custom MCP documentation](https://help.openai.com/en/articles/12584461-connecting-custom-mcp-servers-in-chatgpt) **Other** Use any MCP client that supports remote Streamable HTTP and OAuth. 1. Add a remote MCP server named Lira with this URL: https://lira.thoughtly.com/mcp 2. Choose Streamable HTTP and OAuth, then sign in with your Lira admin account. 3. Ask Lira to show your workspace and available tools. Your client must support remote Streamable HTTP and OAuth. Follow its connection instructions and use the Lira server address above. Setup menus, skill support, and approval controls vary by client. > **Using a private Lira environment?** > > Hosted ChatGPT and Claude connectors need a server address their service can reach. If your workspace is only available on a private network, use Codex or Claude Code on that network, or ask your Lira administrator for a supported hosted connection. ## Add Lira's instructions In **AI agents**, select **Copy instructions** and add them to the instructions your client uses for Lira. If the client supports installable skills, choose **Download skill** instead and follow that client's installation process. You can also [read the current Lira instructions](https://lira.thoughtly.com/mcp/SKILL.md). The instructions help your agent check the workspace, find real scenario and person IDs, preserve unrelated fields, and retry changes safely. They do not replace your client's permission or approval controls. ## Try your first request Start by confirming context: > Show my Lira workspace and the training tools I can use. Then give the agent a specific task, such as: - “Create an unpublished discovery scenario for our onboarding product. Show me the draft before publishing.” - “Summarize training participation and submitted results for the last 30 days.” - “Find the people assigned to this scenario and show their completion status.” Review the workspace and proposed changes before approving a write. Ask the agent to report the outcome and link to the affected item in Lira. A queued generation or ingestion job is still in progress until it finishes. ## What the connection can access Agents can work with scenarios, shared training knowledge, assignments, permitted submitted results, coaching, reports, and certificates. Existing [workspace privacy rules](https://lira.thoughtly.com/docs/workspace/privacy) still apply. Private practice and private preparation are not exposed through an admin connection. Recordings stay protected. Requesting a recording does not create a public sharing link. The connection does not manage member roles or invitations, change workspace privacy, place calls, or issue scores and certificates manually. ## Manage or disconnect an agent Return to **Settings → Integrations → AI agents** to see connection status and last successful use. Copying setup instructions does not create a connection. If Lira is waiting for the first request, return to your agent and ask it to show your workspace, then refresh the connection list. Choose **Disconnect** to stop further requests immediately. Removing a server from your client and disconnecting it in Lira are separate actions. To use a disconnected connection again, choose **Reconnect** in Lira and sign in again if the client asks. ## Troubleshooting | What you see | What to check | | --- | --- | | AI agents is unavailable | Confirm your Lira Admin role. Ask your administrator whether agent connections are enabled. | | Sign-in succeeds but access is denied | Check the selected workspace and your current role. A disconnected connection stays blocked until you reconnect it in Lira. | | The server cannot be reached | Check the address and network access. Hosted clients cannot reach a private or VPN-only server by default. | | A scenario changed while you were editing | Ask the agent to read the latest version before applying your change again. | | A request times out | Ask the agent to check whether the action completed before retrying. Do not repeat a create request with a new retry key. | | A transcript, recording, or score is missing | Check submission status and privacy settings. Missing access is not something the agent can override. | For training setup, see [create a scenario](https://lira.thoughtly.com/docs/manage-training/create-scenarios). For shared results, see [review submitted results](https://lira.thoughtly.com/docs/manage-training/reviewing-results).