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

# Connect Connie to other AI apps

> Set up Connie in Codex, Gemini CLI, Windsurf, Grok, Devin, Cline, opencode and any other app that supports remote MCP with sign-in.

Connie works with any AI app that can connect to a **remote MCP server** and **sign in with OAuth**. This page covers the most common ones. Every app uses the same address, `https://mcp.connie.ai`, and the same Connie sign-in.

**Where in Connie:** click your name at the bottom of the left sidebar → **Connect my agent** · `https://connie.ai/<workspace-id>/connect-agent`. Connie has in-app guides for Codex, Grok, Gemini CLI, Windsurf, Devin, Cline, opencode, OpenClaw, Hermes and **Other MCP clients**.

<Tip>
  After connecting any app, check it works by asking: "List my Connie workspaces."
</Tip>

## Any MCP app (generic setup)

<Steps>
  <Step title="Add a remote server">
    In your app's MCP or connector settings, add a new server. Name it `Connie`.
  </Step>

  <Step title="Use Connie's address and the right transport">
    URL: `https://mcp.connie.ai`. If the app asks for a transport or type, choose **Streamable HTTP** (sometimes just **HTTP** or **Remote**).
  </Step>

  <Step title="Choose OAuth">
    Choose OAuth (sign in) for authentication. Leave client ID and client secret empty. Connie registers your app automatically.
  </Step>

  <Step title="Sign in and allow access">
    Your browser opens Connie. Sign in, choose your **Default workspace** and click **Allow access**.
  </Step>
</Steps>

If your app can only run local (stdio) servers or can't sign in with OAuth, it can't connect to Connie with a sign-in. Some apps can send a fixed header instead; see [API keys](/developers/api-keys).

## Codex

Codex works with a ChatGPT Plus, Pro, Business, Edu or Enterprise plan. The Codex app, CLI and IDE extension share one connection on the same computer.

<Steps>
  <Step title="Install Codex (skip if you have it)">
    Install Node.js from nodejs.org first if you don't have it. Then run `npm install -g @openai/codex`, type `codex`, press Enter and sign in with your ChatGPT account.
  </Step>

  <Step title="Add Connie">
    Run this in Terminal (Mac) or PowerShell (Windows). Codex adds Connie and starts the sign-in straight away.

    ```bash theme={null}
    codex mcp add connie --url https://mcp.connie.ai
    ```

    If your browser doesn't open, hold Cmd (Ctrl on Windows) and click the link Codex shows. If no sign-in starts at all, run `codex mcp login connie`.
  </Step>

  <Step title="Allow access in Connie">
    Choose your **Default workspace** and click **Allow access**. The terminal shows "Successfully logged in."
  </Step>

  <Step title="Use Connie">
    Type `codex`, press Enter and ask away.
  </Step>
</Steps>

* **Codex app or IDE extension instead:** **Settings** → **MCP servers** → **Add server** → **Streamable HTTP**, name `Connie`, URL `https://mcp.connie.ai` → restart → **Authenticate**.
* **Config file** (`~/.codex/config.toml`):
  ```toml theme={null}
  [mcp_servers.connie]
  url = "https://mcp.connie.ai"
  ```
* **Sign in again later:** `codex mcp login connie`. **Check it:** `codex mcp list`.
* Codex cloud tasks need separate setup.
* Connie's in-app Codex guide also has a **Prompt** tab that pastes your personal API key into Codex. We recommend the sign-in steps above instead.

## Gemini CLI

```bash theme={null}
gemini mcp add --transport http --scope user connie https://mcp.connie.ai
```

Start Gemini CLI with `gemini`, then type `/mcp auth connie`. Sign in to Connie in your browser, choose your **Default workspace** and click **Allow access**. Run `/mcp list` to check Connie's tools are there.

Config file (`~/.gemini/settings.json`):

```json settings.json theme={null}
{
  "mcpServers": {
    "connie": { "httpUrl": "https://mcp.connie.ai" }
  }
}
```

<Warning>
  Use `httpUrl`, not `url`. In Gemini CLI, `url` means an older connection type and Connie won't load.
</Warning>

## Windsurf (Devin Desktop)

Windsurf's Cascade agent (now also called Devin Desktop) connects to remote servers with OAuth. In Cascade, click **…** → **Open MCP config file** and add:

```json mcp_config.json theme={null}
{
  "mcpServers": {
    "connie": { "serverUrl": "https://mcp.connie.ai" }
  }
}
```

Save, refresh the MCP servers and complete the Connie sign-in in your browser. Then turn on Connie's tools in Cascade. On Windsurf Team or Enterprise plans, an admin may need to allow MCP servers first.

## Grok

Works on grok.com on any plan. On Business or Enterprise, an admin adds the connector in the xAI console first.

1. Click **Plugins** at the bottom of the sidebar → **New Connector** → **Custom**.
2. Name `Connie`, Server URL `https://mcp.connie.ai` → **Add Connector**.
3. In the Connie window, choose your **Default workspace** and click **Allow access**.
4. In a new chat, click **+** → **Connectors** and switch Connie on.

## Devin

On app.devin.ai, Pro plan or above.

1. **Customize** → **MCPs** → **Add MCP** → **Add custom MCP**.
2. Name `Connie`, Transport **HTTP**, Server URL `https://mcp.connie.ai`, Authentication **OAuth** → **Add**.
3. Tick **I understand and want to proceed** → **Install and enable** → **Connect MCP**.
4. In Connie, click **Allow access**. Start a new session and ask away.

## Cline (in VS Code)

1. Click the Cline icon → the wrench → **MCP** tab → **Add Remote Server**.
2. Name `connie`, URL `https://mcp.connie.ai`, Transport **Streamable HTTP** → **Add Server**.
3. Click **Authenticate**, then **Allow access** in Connie.
4. Connie shows a green dot. Use it in **Act** mode.

## opencode

Run `opencode mcp add`. Enter the name `connie`, choose **Remote**, enter `https://mcp.connie.ai`, answer **Yes** to OAuth and **No** to a pre-registered client ID. Then run `opencode mcp auth connie`, allow access in Connie, and wait for "Authentication successful!".

## OpenClaw and Hermes

Open **Connect my agent** in Connie and choose **OpenClaw** or **Hermes**. The in-app guide gives you the exact commands to copy for your computer.

## Good to know

* **No API key needed** for any app on this page. They all sign in with OAuth.
* **Don't use `mcp-remote` or other local bridges.** Every app on this page connects to Connie directly.
* **Which workspace:** the **Default workspace** you chose, unless you name another in your request. See [Manage access](/mcp/manage-access).
* **Credits:** actions use your Connie workspace's credits. See [Credits](/billing/credits).

## Troubleshooting

<AccordionGroup>
  <Accordion title="Codex was connected, then stopped working and shows Reconnect needed in Connie">
    **Why:** two copies of the same stored sign-in were used, for example after restoring a backup, or when two Codex installs use copies of one login. Connie treats a reused sign-in as a security risk and ends it. **Fix:** run `codex mcp login connie` to sign in again. Don't copy Codex's stored login between installs; if you need two separate setups, add Connie to each one and sign in separately.
  </Accordion>

  <Accordion title="Gemini CLI lists no Connie tools">
    **Fix:** in `settings.json`, use `httpUrl` instead of `url`, then run `/mcp auth connie`.
  </Accordion>

  <Accordion title="My app asks for a client ID or client secret">
    **Fix:** leave them empty, or choose automatic or dynamic registration. Connie doesn't issue client secrets. If the app insists on a secret, it can't connect with a sign-in; see [API keys](/developers/api-keys) for apps that accept a header.
  </Accordion>
</AccordionGroup>

More fixes: [MCP troubleshooting](/mcp/troubleshooting).

## Related

* [Use Connie in your AI app](/mcp/overview)
* [Manage access and workspaces](/mcp/manage-access)
* [API keys](/developers/api-keys)


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