> ## 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 an AI assistant

> Connect FloKit to Claude or ChatGPT over MCP and ask what's going on, what needs your approval, and how your app is set up — in plain language.

FloKit exposes a **Model Context Protocol (MCP)** connector so you can talk to your
growth workspace from Claude or ChatGPT. Connect once, then ask the assistant three
kinds of questions and get answers grounded in your live data — the same numbers the
dashboard shows, never a fabricated figure.

<CardGroup cols={3}>
  <Card title="What's going on?" icon="gauge">
    Health of your data sources, the metrics that matter, and how many approvals are waiting.
  </Card>

  <Card title="Anything I need to approve?" icon="circle-check">
    One unified queue of everything waiting on your decision, oldest first.
  </Card>

  <Card title="Walk me through my app" icon="compass">
    A grounded tour: what's connected, what's live, where revenue comes from, what's gated.
  </Card>
</CardGroup>

<Note>
  The connector is **read-only** today. It can read your workspace and tell you what's
  waiting — it can **never** approve, launch, change, or spend anything. You approve in
  the dashboard. A guarded write path (`decide_approval`) is planned as a separate,
  opt-in scope.
</Note>

## What the assistant can do

Under the hood the connector exposes four tools and three one-click prompts. You don't
call these by name — the assistant picks the right one from your question — but this is
exactly what it can reach:

| Tool                     | Answers                                                | Returns                                                                                                                                                  |
| ------------------------ | ------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `get_status`             | "What's going on?", "How are things?"                  | Source health, KPIs (MRR, active subs, trials, CAC, ROAS — with honest `unavailable` where data is missing), readiness state, count of waiting approvals |
| `list_pending_approvals` | "Anything I need to approve?", "What's waiting on me?" | The unified queue across campaign launches, growth actions, growth-memory drafts, and onboarding go-lives — oldest first, each with a stable id          |
| `get_approval`           | "Tell me about that campaign approval"                 | Full detail for one item: what changes if approved, who requested it, the artifact version a decision pins                                               |
| `explain`                | "Walk me through my app", "Explain my setup"           | A grounded tour built from your overview, integration health, and growth profile. Optional focus: integrations, revenue, growth, campaigns, onboarding   |

**One-click prompts** (shown in the assistant's prompt menu): `daily_briefing`,
`whats_waiting_on_me`, `show_me_around`.

## Example prompts

<CodeGroup>
  ```text Daily check-in theme={null}
  Give me a daily briefing on my app — health, key metrics, and anything waiting on my approval.
  ```

  ```text Approvals theme={null}
  What's waiting for my approval right now? For each one, tell me what it is and your recommendation.
  ```

  ```text Walkthrough theme={null}
  Walk me through my app: what's connected, what's live, and where my revenue comes from.
  ```

  ```text Focused question theme={null}
  What's my MRR and ROAS this month, and is my attribution data healthy?
  ```
</CodeGroup>

## What it can't do

* **Approve, launch, edit, or spend.** Read-only. It surfaces what's waiting and points you to the dashboard to decide.
* **Reach raw ad accounts, payment systems, or the warehouse.** It only sees the same summarized, tenant-scoped reads the dashboard serves.
* **See another company's data.** Your workspace is fixed to the connection's sign-in — it can't be widened by anything you type.

## Security & privacy

<CardGroup cols={2}>
  <Card title="Scoped to you" icon="user-lock">
    Your `companyId` comes from the token you signed in with — never from a prompt. No request can point the connector at another tenant.
  </Card>

  <Card title="Read-only by scope" icon="eye">
    Connections are granted `status:read`. There is no write scope to grant yet, so nothing the assistant reads can trigger an action.
  </Card>

  <Card title="Audited" icon="scroll">
    Reads and (later) writes are attributable to you. Any future approval from chat is recorded in the same hash-chained audit log as the dashboard.
  </Card>

  <Card title="Revocable" icon="power-off">
    Every connection shows up in **Settings → Connected assistants**. Revoke one and it stops working immediately.
  </Card>
</CardGroup>

## Which assistants work

Both Claude and ChatGPT connect to the **same** server — it speaks standard MCP, so
there's nothing vendor-specific to configure.

| Client                                                | How it connects | Guide                                              |
| ----------------------------------------------------- | --------------- | -------------------------------------------------- |
| **Claude.ai** (Pro/Team/Enterprise custom connectors) | OAuth sign-in   | [Connect from Claude](/assistant/connect-claude)   |
| **Claude Code** (CLI)                                 | Bearer token    | [Connect from Claude](/assistant/connect-claude)   |
| **ChatGPT** (Developer Mode connectors)               | OAuth sign-in   | [Connect from ChatGPT](/assistant/connect-chatgpt) |
| **OpenAI API** (Responses / Agents SDK)               | Bearer token    | [Connect from ChatGPT](/assistant/connect-chatgpt) |

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

<CardGroup cols={2}>
  <Card title="Connect from Claude" icon="message" href="/assistant/connect-claude">
    claude.ai custom connector or the Claude Code CLI.
  </Card>

  <Card title="Connect from ChatGPT" icon="robot" href="/assistant/connect-chatgpt">
    ChatGPT Developer Mode or the OpenAI API.
  </Card>
</CardGroup>