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

# ChatGPT Integration

> Give ChatGPT organizational memory and human approval workflows through OrgX MCP.

OrgX connects to ChatGPT through an MCP connection. Use it when ChatGPT needs
durable team memory, decision approvals, artifact recall, agent delegation, or
project execution status.

<Info>
  Connect OrgX from ChatGPT Developer Mode. If Developer Mode is unavailable in
  your account or workspace, use Cursor, Claude, VS Code, or OpenClaw.
</Info>

## Features

<CardGroup cols={2}>
  <Card title="Approve Decisions" icon="check">
    Review and approve pending decisions when the connected ChatGPT client
    supports the OrgX decision widget.
  </Card>

  <Card title="Agent Status" icon="robot">
    See the current agent status exposed by the connected run surface.
  </Card>

  <Card title="Org Memory" icon="brain">
    Search your knowledge graph for context and history.
  </Card>

  <Card title="Spawn Tasks" icon="play">
    Request new work from specialist agents once the connection is
    authenticated.
  </Card>
</CardGroup>

## Getting Started

<Steps>
  <Step title="Enable Developer Mode">
    Open [chatgpt.com/plugins](https://chatgpt.com/plugins). In **Settings →
    Security**, turn **Developer Mode** on.
  </Step>

  {' '}

  <Step title="Add the OrgX reviewer endpoint">
    In Developer Mode, choose **+**, add
    `https://mcp.useorgx.com/mcp?profile=chatgpt`, and save the private Developer
    Mode connection.
  </Step>

  {' '}

  <Step title="Authenticate">
    On first use, open the authentication link, sign in to OrgX, and grant ChatGPT
    permission to access the intended workspace.

    <Info>
      OrgX uses OAuth 2.1 for this connection. Review both OrgX's and the
      connected provider's current data-handling terms.
    </Info>
  </Step>

  <Step title="Start Using">
    Once connected, just talk to ChatGPT naturally:

    * *"Show me pending decisions"*
    * *"What are my agents working on?"*
    * *"Approve the marketing campaign decision"*
    * *"Search for our pricing strategy decisions"*
    * *"Remember this decision for the team: we are moving onboarding analytics to PostHog."*
  </Step>
</Steps>

## Available Commands

### Tier-0 Intent Tools

Start with these wrapper tools. They map to the natural phrases ChatGPT users
actually type while preserving the underlying OrgX behavior.

| User intent               | Preferred tool   | Underlying OrgX action                                                |
| ------------------------- | ---------------- | --------------------------------------------------------------------- |
| Remember a team decision  | `orgx_decide`    | Records a decision in organizational memory                           |
| Recall prior context      | `orgx_search`    | Searches decisions, artifacts, initiatives, and project context       |
| Review blocked agent work | `orgx_decide`    | Lists pending approvals, then approves or rejects after confirmation  |
| Hand work to an agent     | `orgx_spawn`     | Assigns work to a specialist OrgX agent                               |
| Check project progress    | `orgx_recommend` | Returns initiative health, blockers, milestones, owners, and activity |

### Decision Management

<AccordionGroup>
  <Accordion title="Get Pending Decisions" icon="list">
    **What it does**: Retrieves all decisions waiting for your approval.

    **Example prompts**:

    * "Show my pending decisions"
    * "What needs my approval?"
    * "Any critical decisions waiting?"

    **Parameters**:

    * `limit` (optional): Maximum number to return
    * `urgency_filter` (optional): `all`, `critical`, or `high`
    * `initiative_id` (optional): Filter by specific initiative

    **Response**: The client may render decision cards; response fields depend
    on the connected surface.
  </Accordion>

  <Accordion title="Approve Decision" icon="check">
    **What it does**: Records approval for a pending decision. A configured
    follow-on action may then become eligible or be dispatched; approval alone
    is not proof of an external result.

    **Example prompts**:

    * "Approve the homepage redesign decision"
    * "Approve decision \[decision\_id] with note: approved after review"

    **Parameters**:

    * `decision_id` (required): ID of the decision
    * `note` (optional): Comment to attach

    **Response**: The client returns the fields available for the recorded
    decision. Inspect the record and any downstream provider receipt separately.
  </Accordion>

  <Accordion title="Reject Decision" icon="xmark">
    **What it does**: Rejects a decision with feedback for the agent.

    **Example prompts**:

    * "Reject the email campaign, needs more personalization"
    * "Reject decision \[decision\_id] because the budget is too high"

    **Parameters**:

    * `decision_id` (required): ID of the decision
    * `reason` (required): Why you're rejecting

    **Response**: The client returns the fields available for the recorded
    decision. A configured workflow may route the feedback for revision.
  </Accordion>

  <Accordion title="Get Decision History" icon="clock-rotate-left">
    **What it does**: Retrieves past decisions on a topic.

    **Example prompts**:

    * "What decisions have we made about pricing?"
    * "Show me past marketing decisions"

    **Parameters**:

    * `topic` (required): What to search for
    * `initiative_id` (optional): Filter by initiative
    * `limit` (optional): Maximum results
  </Accordion>
</AccordionGroup>

### Agent Management

<AccordionGroup>
  <Accordion title="Get Agent Status" icon="robot">
    **What it does**: Shows current agent activity and status.

    **Example prompts**:

    * "What are my agents doing?"
    * "Show me active agents"
    * "Is the engineering agent working on anything?"

    **Parameters**:

    * `agent_id` (optional): Filter to specific agent
    * `include_idle` (optional): Show idle agents too

    **Response**: The client may render agent cards with the current fields
    returned by the connected run surface.
  </Accordion>

  <Accordion title="Spawn Agent Task" icon="play">
    **What it does**: Assigns a new task to a specialist agent.

    **Example prompts**:

    * "Have the product agent analyze our competitor positioning"
    * "Ask the marketing agent to draft a blog post about our new feature"

    **Parameters**:

    * `agent` (required): Canonical agent ID (`engineering-agent`, `product-agent`, `marketing-agent`, `sales-agent`, `design-agent`, `operations-agent`, `orchestrator-agent`)
    * `task` (required): What to do
    * `context` (optional): Additional background
    * `initiative_id` (optional): Link to an initiative

    **Response**: The client returns the accepted, blocked, or recorded state
    available for the delegation request. Do not infer completion from dispatch.

    <Warning>
      This writes to your workspace. A connection that is authenticated can
      call it — the `agents:write` scope is recorded at consent time but is not
      checked on the call. Authorize ChatGPT from an account whose workspace
      membership matches the writes you want it to be able to make.
    </Warning>
  </Accordion>
</AccordionGroup>

### Memory & Search

<AccordionGroup>
  <Accordion title="Query Org Memory" icon="brain">
    **What it does**: Searches your organizational knowledge graph.

    **Example prompts**:

    * "What do we know about enterprise pricing?"
    * "Find artifacts related to \[initiative\_or\_topic]"
    * "Search for previous security audits"

    **Parameters**:

    * `query` (required): What to search for
    * `scope` (optional): `all`, `artifacts`, `decisions`, or `initiatives`
    * `limit` (optional): Maximum results

    **Response**: The client may render relevant items with excerpts and links.
  </Accordion>

  <Accordion title="Get Initiative Pulse" icon="heart-pulse">
    **What it does**: Shows health and progress of an initiative.

    **Example prompts**:

    * "How is the product launch initiative doing?"
    * "Show me the pulse for initiative \[initiative\_id]"

    **Parameters**:

    * `initiative_id` (required): Which initiative to check

    **Response**: The client may render the initiative fields returned by the
    connected OrgX surface.
  </Accordion>
</AccordionGroup>

## Widgets

When the connected ChatGPT surface supports them, OrgX can use rich **widgets**
to display information rather than plain text:

<CardGroup cols={2}>
  <Card title="Decisions Widget" icon="list-check">
    Interactive cards with approve/reject buttons, context, and evidence links.
  </Card>

  <Card title="Agent Status Widget" icon="robot">
    Visual cards showing each agent's current task, progress, and health.
  </Card>

  <Card title="Search Results Widget" icon="magnifying-glass">
    Organized results with type icons, excerpts, and deep links.
  </Card>

  <Card title="Initiative Pulse Widget" icon="chart-line">
    Dashboard view with health score, milestone progress, and activity feed.
  </Card>
</CardGroup>

## Permissions & Security

### Scopes

The consent screen lists the scopes the connection declares. They describe
intent and appear back in the `orgx_bootstrap` payload as `granted_scopes`.

| Scope               | What it declares                |
| ------------------- | ------------------------------- |
| `decisions:read`    | View pending and past decisions |
| `decisions:write`   | Approve or reject decisions     |
| `agents:read`       | View agent status and activity  |
| `agents:write`      | Spawn new agent tasks           |
| `initiatives:read`  | View initiative details         |
| `initiatives:write` | Create and update initiatives   |
| `memory:read`       | Search organizational knowledge |
| `offline_access`    | Issue a refresh token           |

<Warning>
  **Scopes are consent and discovery metadata, not a privilege boundary.** OrgX
  does not compare a tool call against the scopes you granted. What is enforced
  is authentication — an OAuth-declaring tool is refused without a signed-in
  user — and workspace membership, which decides which records that user can
  reach. Approving fewer scopes does not make the connection read-only. Connect
  ChatGPT from an account whose workspace access you are willing to give it, and
  see [Declared scopes](/docs/api/mcp-protocol#declared-scopes) for the mechanism.
</Warning>

<Info>
  The authenticated ChatGPT reviewer endpoint is
  `https://mcp.useorgx.com/mcp?profile=chatgpt` and requires OrgX OAuth. For
  no-auth exploration, use the discovery-only public endpoint at
  `https://mcp.useorgx.com/public`; it returns examples and connection help, not
  workspace data.
</Info>

### Data Privacy

* Your messages in ChatGPT are processed by OpenAI per their policies
* OrgX receives the MCP requests and responses required by the connection, not
  an implied copy of the entire ChatGPT conversation
* All API calls are authenticated and scoped to your workspace
* Data handling and retention follow the current OrgX privacy policy and the
  connected provider's policy; this page does not promise zero retention

## Troubleshooting

<AccordionGroup>
  <Accordion title="App not appearing in ChatGPT">
    1. Verify **Settings -> Security -> Developer Mode** is enabled
    2. Confirm you've added `https://mcp.useorgx.com/mcp?profile=chatgpt`
    3. Restart ChatGPT if the connection does not appear
  </Accordion>

  <Accordion title="Authentication failing">
    1. Try clearing your browser cache and re-authenticating 2. Make sure you're
       signing into the correct OrgX workspace 3. Check that your OrgX account is in
       good standing
  </Accordion>

  <Accordion title="Decisions not loading">
    * Ensure you have pending decisions in your OrgX workspace - Confirm the
      connection is signed in and pointed at the workspace that holds them - Try
      refreshing by asking "Show pending decisions" again
  </Accordion>

  <Accordion title="Can't approve/reject decisions">
    Write operations require a signed-in connection whose account can act in
    that workspace. Re-authenticate, and confirm the account you signed in with
    is a member of the workspace holding the decision.
  </Accordion>
</AccordionGroup>

## Next Steps

<CardGroup cols={2}>
  <Card title="MCP Client Setup" icon="code" href="/docs/guides/cursor-mcp-setup">
    Connect OrgX from Cursor, Claude, VS Code, and other MCP clients.
  </Card>

  <Card title="MCP Tools Reference" icon="server" href="/docs/api/mcp-tools">
    Current reference for the published MCP tools.
  </Card>
</CardGroup>