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

# Connect your AI app

> Link ChatGPT, Claude, Claude Code, Cursor, or any MCP-capable app to Agentry.

There are two ways to connect, matching the two sections of the Console's Agentry page:

* **Via AI app** — install Agentry from your AI app's own directory and authorize it with a **connect code**. No URLs or secrets to copy. Best for ChatGPT and Claude users.
* **Via MCP** — point an MCP client (Claude Code, Cursor, or any custom MCP-capable app) at the Agentry MCP server and authenticate with a **connector credential** (client ID + secret).

Either way, your administrator provides what you need from **Console → Settings → Agentry** — see [Manage access](/ai/agentry/manage-access).

## Via AI app

<Frame>
  <img src="https://mintcdn.com/social-b97141fb/-t88VJFqwt-Dt9V9/images/ai/agentry-connect-flow.svg?fit=max&auto=format&n=-t88VJFqwt-Dt9V9&q=85&s=b451e6602e4cc7b413f9277257b26581" alt="Three steps: install Social Plus Agentry from your AI app's directory, authorize with a connect code, then ask" width="760" height="210" data-path="images/ai/agentry-connect-flow.svg" />
</Frame>

<Tabs>
  <Tab title="Claude">
    <Steps>
      <Step title="Get a connect code">
        Ask your admin for a connect code, or generate one yourself in Console → Settings → Agentry (super admin). Codes are single-use and short-lived, so get one just before you connect.
      </Step>

      <Step title="Find Agentry in Claude">
        In Claude, open **Customize** in the sidebar and choose the **Connectors** tab. Select **+**, choose **Browse connectors**, and search for **Social Plus Agentry**.
      </Step>

      <Step title="Connect and authorize">
        Choose **Connect**. A social.plus page asks for your connect code — paste it and approve access.
      </Step>

      <Step title="Ask">
        Start a chat and ask about your data. Claude calls Agentry and answers from your first-party data.
      </Step>
    </Steps>
  </Tab>

  <Tab title="ChatGPT">
    Coming soon..
  </Tab>
</Tabs>

## Via MCP

Each guide uses the same **MCP server URL** (shown on the Console's Agentry page) plus a connector credential your admin creates there. The credential type must match your client — the create dialog offers **Claude (web connector)**, **Native / CLI**, and **Custom redirect URL**.

<Tabs>
  <Tab title="Claude (custom connector)">
    <Steps>
      <Step title="Create a credential">
        In the Console, create a **Claude (web connector)** credential and copy its Client ID and Client Secret — the secret is shown only once.
      </Step>

      <Step title="Add the connector">
        In Claude, open **Customize → Connectors** (Organization settings → Connectors on Team and Enterprise plans), select **+**, and choose **Add custom connector**.
      </Step>

      <Step title="Enter the server details">
        Paste the MCP server URL. Open **Advanced settings** and paste the Client ID and Client Secret.
      </Step>

      <Step title="Connect">
        Choose **Add**, then **Connect**, and approve access.
      </Step>
    </Steps>
  </Tab>

  <Tab title="Claude Code (CLI)">
    <Steps>
      <Step title="Create a credential">
        In the Console, create a **Native / CLI** credential and copy its Client ID and Client Secret.
      </Step>

      <Step title="Add the server">
        The Console's Claude Code tab shows the exact command with the MCP URL and your Client ID filled in. The bare `--client-secret` flag prompts for the secret (or reads `MCP_CLIENT_SECRET` from the environment).
      </Step>

      <Step title="Approve">
        On first connect, approve access in the browser tab that opens.
      </Step>
    </Steps>
  </Tab>

  <Tab title="Cursor">
    <Steps>
      <Step title="Create a credential">
        In the Console, create a **Native / CLI** credential and copy its Client ID and Client Secret.
      </Step>

      <Step title="Add the server">
        Add Agentry to your Cursor `mcp.json` using the snippet shown on the Console's Cursor tab, filling in the Client ID and Client Secret. Cursor also supports referencing environment variables with its `env:` prefix instead of pasting the secret.
      </Step>

      <Step title="Approve">
        On first connect, approve access in the browser tab Cursor opens.
      </Step>
    </Steps>
  </Tab>

  <Tab title="Any other MCP client">
    Any MCP-capable app connects the same way: point it at the MCP server URL and authenticate with a connector credential. This is also the right path for **AI agents in agentic workflows** — the credential outlives any one session, so after a one-time authorization approval, an autonomous agent keeps its access through normal token refresh.

    <Steps>
      <Step title="Create a credential">
        For an app that runs its own OAuth callback, your admin should create the credential with a **Custom redirect URL**; local and CLI tools use a **Native / CLI** credential (loopback redirect).
      </Step>

      <Step title="Add the server">
        In your AI app, open its MCP or custom-connector settings, add a new server, and paste the MCP server URL.
      </Step>

      <Step title="Authenticate">
        When prompted, authenticate with the Client ID and Client Secret and approve access. Your app then discovers Agentry's tools and can ask about your first-party data.
      </Step>
    </Steps>
  </Tab>
</Tabs>

## Troubleshooting

* **The connect code is rejected.** Codes are single-use and expire — generate a fresh one and retry.
* **The app connected but answers stopped.** An admin may have removed the connection; reconnect with a new code, or a new sign-in for credential-based clients.
* **Wrong or empty answers about old activity.** Questions beyond the roughly 90-day data window can't be answered from Agentry's data.