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

# Connect from ChatGPT

> Add FloKit to ChatGPT as a Developer Mode connector, or call it from the OpenAI API with a token.

FloKit connects to OpenAI two ways: as a **connector in ChatGPT** (sign in with OAuth),
or as an **MCP tool from the OpenAI API** (Responses API or Agents SDK) with an access
token. Same read-only server, same tools.

<Card title="Server URL" icon="link" horizontal>
  `https://api.dashboard.flokitai.com/mcp`
</Card>

## Option A — ChatGPT Developer Mode connector

<Note>
  Custom MCP connectors live under **Developer Mode**, available on ChatGPT Plus, Pro,
  Business, and Enterprise. Enable it in **Settings → Connectors → Advanced → Developer
  mode**.
</Note>

<Steps>
  <Step title="Create a new connector">
    In **Settings → Connectors**, choose **New connector** (Developer Mode). Name it
    **FloKit** and paste the server URL, with authentication set to **OAuth**.

    <Frame caption="ChatGPT → Settings → Connectors → New connector">
      <img src="https://mintcdn.com/flokitai/dfSJ5h3oFVLOkPx8/images/assistant/chatgpt-add-connector.svg?fit=max&auto=format&n=dfSJ5h3oFVLOkPx8&q=85&s=5bc5a94e14bafd309e9acfd80981df15" alt="ChatGPT new connector dialog with the FloKit MCP URL and OAuth" width="860" height="520" data-path="images/assistant/chatgpt-add-connector.svg" />
    </Frame>
  </Step>

  <Step title="Sign in and approve">
    ChatGPT opens the FloKit consent screen. Sign in with your FloKit work email and click
    **Allow access** — the screen confirms read-only access.

    <Frame caption="The FloKit consent screen — read-only access">
      <img src="https://mintcdn.com/flokitai/dfSJ5h3oFVLOkPx8/images/assistant/consent.svg?fit=max&auto=format&n=dfSJ5h3oFVLOkPx8&q=85&s=d75a5628f85225cc99130ee5e58a31c1" alt="FloKit OAuth consent screen granting status:read" width="860" height="520" data-path="images/assistant/consent.svg" />
    </Frame>
  </Step>

  <Step title="Ask FloKit something">
    Start a chat with the FloKit connector enabled and try:

    ```text theme={null}
    What's waiting for my approval, and what's my MRR this month?
    ```
  </Step>
</Steps>

## Option B — OpenAI API (Responses / Agents)

Point the OpenAI **Responses API** MCP tool at FloKit and pass your access token as the
`authorization` value — no browser sign-in.

<Steps>
  <Step title="Create an access token in FloKit">
    In the dashboard, **Settings → Connected assistants → Create**. Copy the token (starts
    with `flk_live_`).

    <Frame caption="Dashboard → Settings → Connected assistants">
      <img src="https://mintcdn.com/flokitai/dfSJ5h3oFVLOkPx8/images/assistant/dashboard-connect.svg?fit=max&auto=format&n=dfSJ5h3oFVLOkPx8&q=85&s=d3f0da436abb8a0412517162de50c1a4" alt="FloKit dashboard connected assistants panel with server URL and access token" width="860" height="520" data-path="images/assistant/dashboard-connect.svg" />
    </Frame>
  </Step>

  <Step title="Call it as an MCP tool">
    <CodeGroup>
      ```python Python theme={null}
      from openai import OpenAI

      client = OpenAI()

      resp = client.responses.create(
          model="gpt-5",  # any model that supports MCP tools
          tools=[{
              "type": "mcp",
              "server_label": "flokit",
              "server_url": "https://api.dashboard.flokitai.com/mcp",
              "authorization": "flk_live_your_token_here",
              "require_approval": "never",  # all FloKit tools are read-only
          }],
          input="What's going on with my app, and is anything waiting on my approval?",
      )

      print(resp.output_text)
      ```

      ```typescript TypeScript theme={null}
      import OpenAI from "openai";

      const client = new OpenAI();

      const resp = await client.responses.create({
        model: "gpt-5", // any model that supports MCP tools
        tools: [{
          type: "mcp",
          server_label: "flokit",
          server_url: "https://api.dashboard.flokitai.com/mcp",
          authorization: "flk_live_your_token_here",
          require_approval: "never", // all FloKit tools are read-only
        }],
        input: "What's going on with my app, and is anything waiting on my approval?",
      });

      console.log(resp.output_text);
      ```
    </CodeGroup>
  </Step>
</Steps>

<Tip>
  Because every FloKit tool is read-only, `require_approval: "never"` is safe — the model
  can't trigger a side effect. Keep the token server-side; never ship it to a browser or
  mobile client.
</Tip>

## Troubleshooting

<AccordionGroup>
  <Accordion title="401 / authorization error from the tool">
    The token was revoked or expired. Mint a fresh one in **Settings → Connected
    assistants**.
  </Accordion>

  <Accordion title="The model won't call the tool">
    Make sure your prompt asks for something FloKit knows — status, metrics, or approvals.
    Vague prompts may not trigger a tool call; try one of the [example prompts](/assistant/overview#example-prompts).
  </Accordion>
</AccordionGroup>