# AI Integrations

How to connect SiteTrax.io Intelligence to the AI assistant your team already uses, and what that assistant can and cannot reach.

This book covers connecting ChatGPT and Claude step by step, what the Model Context Protocol is and why SiteTrax.io implements it, how to phrase operational questions so they return usable answers, which platforms are supported today, and how to triage a connection that is not working.

The SiteTrax.io MCP Server book on this shelf holds the full technical contract, including transport, scopes and response semantics.

# Connecting ChatGPT

![Connecting ChatGPT](https://docs.sitetrax.io/uploads/images/gallery/2026-09/st-intel-connect-chatgpt-feature.png)

SiteTrax.io publishes an app for ChatGPT. Connecting it lets you ask operational questions in ChatGPT and get answers drawn from the SiteTrax.io projects your account is permitted to see.

## Before you start

- A SiteTrax.io account with at least one shared project.
- A ChatGPT account. On workspace plans, an administrator controls whether apps may be added and how they are rolled out.
- A desktop or laptop browser. Apps built on the Model Context Protocol are supported on the web, not in the ChatGPT mobile apps.

## Connect from SiteTrax.io

This is the shortest path and it is the one to use unless your organization requires its own configuration.

1. Sign in at [app.sitetrax.io](https://app.sitetrax.io) and open **Intelligence**.
2. Scroll to **Use SiteTrax.io with your assistant**.
3. On the ChatGPT card, select **Open in ChatGPT**. It opens in a new tab.
4. When prompted, sign in to SiteTrax.io and review the permissions being requested.
5. Authorize the connection, then return to ChatGPT and confirm SiteTrax.io is available.

![The assistant connection cards on the Intelligence hub](https://docs.sitetrax.io/uploads/images/gallery/2026-09/st-intel-ui-assistant-connections.png)

Authentication is completed through the SiteTrax.io authorization screen. Never type a SiteTrax.io password into an app configuration field, and never paste access tokens into a prompt.

## Configuring your own connection

Organizations that manage their own apps can point ChatGPT at the SiteTrax.io remote MCP endpoint instead.

Endpoint: `https://mcp.sitetrax.io/mcp`

OpenAI documents this as developer mode. In outline: an authorized user enables developer mode under app settings, creates the app with the endpoint and its metadata, selects the authentication method, scans the available tools, completes the authorization prompt, and creates the app. Availability differs by plan, with administrators controlling access on Business, Enterprise and Edu, and Pro accounts limited to read and fetch behaviour. Follow OpenAI current instructions rather than a fixed set of screens, because this interface changes.

## Check the connection

Ask something you can verify by hand, and name SiteTrax.io in the prompt.

> Using SiteTrax.io, list the projects I have access to.

If projects come back that match what you see in the application, the connection is working and scoped correctly.

## What ChatGPT can reach

Only what your SiteTrax.io account can reach. Connecting an app does not widen SiteTrax.io permissions. See [Using MCP with SiteTrax.io](/books/ai-integrations/page/using-mcp-with-sitetraxio) for the capability groups, and the [SiteTrax.io MCP Server](/books/mcp-server/page/sitetraxio-mcp-server) reference for the full technical contract.

<p class="callout warning">If answers come back generic rather than tailored to your operation, check that you authorized with the same SiteTrax.io account that published your Intelligence profile. See [Troubleshooting Connections](/books/ai-integrations/page/troubleshooting-connections).</p>

## Sources

- [Developer mode and MCP apps in ChatGPT, OpenAI Help Center](https://help.openai.com/en/articles/12584461-developer-mode-and-mcp-apps-in-chatgpt)
- [Apps in ChatGPT, OpenAI Help Center](https://help.openai.com/en/articles/11487775-connectors-in-chatgpt)

# Connecting Claude

![Connecting Claude](https://docs.sitetrax.io/uploads/images/gallery/2026-09/st-intel-connect-claude-feature.png)

SiteTrax.io connects to Claude as a custom connector over the remote MCP endpoint. Once it is added, you can ask operational questions in Claude and get answers scoped to your SiteTrax.io projects.

## Before you start

- A SiteTrax.io account with at least one shared project.
- A Claude account on a plan that supports custom connectors. Anthropic documents these as Pro, Max, Team and Enterprise.
- On Team and Enterprise, only Owners can add a custom connector. Everyone else connects to one an Owner has already added.

## Connect from SiteTrax.io

1. Sign in at [app.sitetrax.io](https://app.sitetrax.io) and open **Intelligence**.
2. Scroll to **Use SiteTrax.io with your assistant**.
3. On the Claude card, select **Add to Claude**. It opens Claude connector settings in a new tab with the SiteTrax.io connector name and URL already filled in.
4. Confirm the details and add the connector.
5. Select **Connect**, sign in to SiteTrax.io when prompted, review the requested permissions and authorize.

![The Claude connection card on the Intelligence hub](https://docs.sitetrax.io/uploads/images/gallery/2026-09/st-intel-ui-assistant-connections.png)

## Adding it manually

If you would rather add it by hand, or the deep link is blocked by your browser:

1. Open Claude settings and go to **Connectors**.
2. Select **Add custom connector**.
3. Enter the SiteTrax.io remote MCP server URL: `https://mcp.sitetrax.io/mcp`
4. Leave the advanced OAuth client fields empty unless SiteTrax.io has given you values for them.
5. Add the connector, then select **Connect** and complete the SiteTrax.io sign-in.

## Turn it on in a conversation

Adding a connector makes it available; it is enabled per conversation. Use the plus button in the message box, choose connectors, and switch SiteTrax.io on for that conversation.

## Check the connection

> Using SiteTrax.io, list the projects I have access to.

Projects matching what you see in the application confirm the connection and its scope.

## Permissions

Standard operational queries use read access. Notification and digest capabilities require a separate write scope and should be approved only when your workflow needs them. The [SiteTrax.io MCP Server](/books/mcp-server/page/sitetraxio-mcp-server) reference covers scopes in full.

<p class="callout warning">Sign in to the connector with the same SiteTrax.io account that published your Intelligence profile. A different account authenticates successfully but carries different project access and no profile context.</p>

## Sources

- [Getting started with custom connectors using remote MCP, Anthropic Support](https://support.claude.com/en/articles/11175166-getting-started-with-custom-connectors-using-remote-mcp)

# Understanding MCP

![Understanding MCP](https://docs.sitetrax.io/uploads/images/gallery/2026-09/st-intel-understanding-mcp-feature.png)

Model Context Protocol, usually shortened to MCP, is an open standard that lets an AI assistant connect to an outside system and use it. SiteTrax.io implements it so that assistants can work with your operational data without anyone exporting it first.

## Why it exists

Before MCP, getting an assistant to answer a question about your yard meant copying data into a prompt or building a bespoke integration for each assistant. Both approaches age badly. A copied extract is stale the moment it is pasted, and a bespoke integration has to be rebuilt for the next tool.

MCP replaces that with one connection the assistant understands. SiteTrax.io publishes what it can do once, and any compatible assistant can use it.

## What it means in practice

<table id="bkmrk-without-mcpwith-mcpe"><thead><tr><th>Without MCP</th><th>With MCP</th></tr></thead><tbody><tr><td>Export a report, paste it into a chat</td><td>The assistant reads the live record at the moment you ask</td></tr><tr><td>The extract reflects one moment and one filter</td><td>Every question is answered against current data</td></tr><tr><td>The assistant cannot check its own answer</td><td>The assistant can retrieve the supporting record and its image</td></tr><tr><td>Sharing the extract shares everything in it</td><td>Access follows the signed-in SiteTrax.io account</td></tr></tbody></table>

## How SiteTrax.io implements it

- **It is a remote server.** The endpoint is `https://mcp.sitetrax.io/mcp`. An assistant that only supports servers running on your own machine cannot reach it.
- **Authorization is OAuth-based.** You sign in to SiteTrax.io through its own authorization screen. No SiteTrax.io password is entered into the assistant.
- **Permissions are scoped.** Read access covers operational queries. Notification and digest actions need a separate write scope that is approved separately.
- **Analysis happens on the server.** SiteTrax.io calculates totals, groupings and durations itself, so the assistant does not pull down raw records to do arithmetic.
- **SiteTrax.io stays the system of record.** The assistant reads; it does not become the source of truth.

## What it does not do

- It does not widen your SiteTrax.io permissions. A project not shared with you stays out of reach.
- It does not hand your data to the assistant vendor as a bulk export.
- It does not make an unverified answer verified. Interpretation rules still apply.

For the full technical contract, including transport, scopes and response semantics, see the [SiteTrax.io MCP Server](/books/mcp-server/page/sitetraxio-mcp-server) reference.

# Using MCP with SiteTrax.io

![Using MCP with SiteTrax.io](https://docs.sitetrax.io/uploads/images/gallery/2026-09/st-intel-using-mcp-feature.png)

Once an assistant is connected, the skill is asking well. This page covers what to ask for and how to phrase it. For scopes, tool names and response contracts, see the [SiteTrax.io MCP Server](/books/mcp-server/page/sitetraxio-mcp-server) reference.

## What you can ask for

<table id="bkmrk-if-you-need-toask-fo"><thead><tr><th>If you need to</th><th>Ask for</th></tr></thead><tbody><tr><td>Locate one unit</td><td>The most recent observation of a specific identifier, with project, time and image</td></tr><tr><td>Understand a period</td><td>Activity across your sites for a date range, with totals and trends</td></tr><tr><td>Compare</td><td>A breakdown by project, camera, asset type, status code or time period</td></tr><tr><td>Measure dwell</td><td>Observed time between defined operational boundaries, such as gate-in to gate-out</td></tr><tr><td>Find problems</td><td>Records meeting review criteria, unreadable results, or unusual volume gaps</td></tr><tr><td>Check capture health</td><td>Camera activity and scan health across your projects</td></tr><tr><td>Get the evidence</td><td>The detection image or video metadata behind a specific record</td></tr></tbody></table>

## How to phrase it

Six things make the difference between a vague answer and a usable one.

1. **Name SiteTrax.io** so the assistant reaches for the connector rather than answering from memory.
2. **Name the site or project.** Without it, the answer spans everything you can see.
3. **Give a date or time range.** This is the single most common cause of a surprising number.
4. **Say the asset type** when it matters, such as containers rather than chassis.
5. **Use the complete identifier** when looking for one unit. A partial identifier will not match.
6. **Ask for the supporting record** whenever the answer will be used in an invoice, a dispute or a report.

### A weak prompt

> How busy were we?

### The same question, answerable

> Using SiteTrax.io, summarize container and chassis activity at the Utah site for the last seven days and break the results down by status code and asset type.

## Reading the answer honestly

These are the mistakes that turn a good answer into a wrong decision.

- **A record is a stored processing result**, not a guaranteed physical event. Combined container and chassis processing can add a synthetic row for the paired side.
- **A processing status is not a condition report.** An unreadable result does not establish that the unit is damaged.
- **An observed interval is not an explanation.** Time between two camera observations is a measurement, not a cause.
- **A USDOT number identifies a carrier**, not an individual tractor.
- **Missing pairing is reported, not estimated.** Where a sequence cannot be paired reliably, SiteTrax.io says so instead of filling the gap.

## When the answer will be acted on

Ask for the record. SiteTrax.io can return a link in the form `app.sitetrax.io/dashboard/assets/view/{asset_id}`, which opens the same detail page you would reach in the application, with the image and the full payload. Verify there before an invoice is disputed or a charge is paid.

![The asset record an assistant answer can be traced back to](https://docs.sitetrax.io/uploads/images/gallery/2026-09/st-intel-ui-asset-detail.png)

<p class="callout info">Your published Intelligence profile shapes the response style you get back. If you asked for answer plus why, expect a short explanation underneath the number.</p>

# Supported AI Platforms

![Supported AI Platforms](https://docs.sitetrax.io/uploads/images/gallery/2026-09/st-intel-supported-platforms-feature.png)

SiteTrax.io exposes one remote MCP endpoint, so any assistant that supports remote MCP servers with OAuth can use it. The table below records the platforms SiteTrax.io provides a direct route for, and their current status in the application.

## Status

<table id="bkmrk-platformhow-it-conne"><thead><tr><th>Platform</th><th>How it connects</th><th>Status</th><th>Guide</th></tr></thead><tbody><tr><td>ChatGPT</td><td>Published SiteTrax.io app, opened from the Intelligence hub</td><td>Available</td><td>[Connecting ChatGPT](/books/ai-integrations/page/connecting-chatgpt)</td></tr><tr><td>Claude</td><td>Custom connector pointed at the SiteTrax.io remote MCP endpoint</td><td>Available</td><td>[Connecting Claude](/books/ai-integrations/page/connecting-claude)</td></tr><tr><td>Microsoft 365 Copilot</td><td>SiteTrax.io connector</td><td>Not yet available. The Intelligence hub lists Copilot for visibility and states that the connection is still being built.</td><td>[Connect SiteTrax.io to Microsoft 365 Copilot](/books/mcp-server/page/connect-sitetraxio-to-microsoft-365-copilot)</td></tr><tr><td>Other MCP-compatible clients</td><td>Add `https://mcp.sitetrax.io/mcp` as a remote MCP server</td><td>Supported where the client meets the requirements below</td><td>[SiteTrax.io MCP Server](/books/mcp-server/page/sitetraxio-mcp-server)</td></tr></tbody></table>

<p class="callout warning">The Microsoft 365 Copilot guide describes the connection as designed. Until the Intelligence hub stops showing Copilot as coming soon, treat that page as preparation rather than as a connection you can complete today.</p>

## What a client needs to qualify

- Support for **remote** MCP servers. A client that only runs local servers on your own machine cannot reach the endpoint.
- **Streamable HTTP** transport. Server-Sent Events transport is not supported.
- **OAuth** support, so authorization completes on the SiteTrax.io side.
- Support for MCP **tools and resources**. SiteTrax.io does not expose MCP prompts.

## Platform limits worth knowing before you commit

- **ChatGPT.** Apps built on MCP are supported on the web rather than in the mobile apps. On workspace plans an administrator controls whether apps can be added.
- **Claude.** Custom connectors are available on Pro, Max, Team and Enterprise. On Team and Enterprise only Owners can add one; other members connect afterwards.

## Choosing between them

There is no data difference. Every platform reaches the same records under the same permissions, and analytical work happens on the SiteTrax.io server either way. Choose the assistant your team already uses, and connect with the SiteTrax.io account that holds your project access and published profile.

# Troubleshooting Connections

![Troubleshooting Connections](https://docs.sitetrax.io/uploads/images/gallery/2026-09/st-intel-troubleshooting-connections-feature.png)

Connection problems, in the order they usually appear. For problems with the data in an answer rather than the connection itself, see the troubleshooting section of the [SiteTrax.io MCP Server](/books/mcp-server/page/sitetraxio-mcp-server) reference.

## Setting up

<table id="bkmrk-what-you-seelikely-c"><thead><tr><th>What you see</th><th>Likely cause</th><th>What to do</th></tr></thead><tbody><tr><td>The assistant cannot reach the server at all</td><td>The client does not support remote MCP servers, or a network control is blocking the endpoint</td><td>Confirm the endpoint is `https://mcp.sitetrax.io/mcp`, that the client supports remote servers over HTTPS with OAuth, and that no firewall or security tool blocks it</td></tr><tr><td>Nothing happens when authorization should start</td><td>The client only supports local servers running on your own machine</td><td>A local-only client cannot reach a remote endpoint. Use a client that supports remote MCP servers</td></tr><tr><td>There is no option to add a connector</td><td>Plan or role restriction</td><td>On Claude Team and Enterprise only Owners can add custom connectors. In ChatGPT workspaces an administrator controls app creation. Ask whoever administers your assistant</td></tr><tr><td>The Add to Claude or Open in ChatGPT link does nothing</td><td>The new tab was blocked</td><td>Allow the pop-up, or add the connector manually using the endpoint above</td></tr><tr><td>Microsoft Copilot is shown but cannot be selected</td><td>That connection is still being built</td><td>Use ChatGPT or Claude for now. See [Supported AI Platforms](/books/ai-integrations/page/supported-ai-platforms)</td></tr><tr><td>It works on a laptop but not on a phone</td><td>Platform limitation</td><td>ChatGPT apps built on MCP are supported on the web rather than in the mobile apps</td></tr></tbody></table>

## After connecting

<table id="bkmrk-what-you-seelikely-c-1"><thead><tr><th>What you see</th><th>Likely cause</th><th>What to do</th></tr></thead><tbody><tr><td>Authorization succeeded but no projects appear</td><td>No projects are shared with the account you used, or you signed in with a different account</td><td>Confirm which account you authorized with, and that at least one project is shared with it. This is a sharing question rather than an authentication one</td></tr><tr><td>Answers are generic and ignore your priorities</td><td>The account connected to the assistant is not the account that published your Intelligence profile</td><td>Profiles are held per user. Reconnect using the same SiteTrax.io account you used to publish, then ask again</td></tr><tr><td>Fewer projects than you expect</td><td>The assistant is scoped to a different account with narrower access</td><td>Compare the project list in the assistant against Projects in the application while signed in as the same user</td></tr><tr><td>A tool reports that another permission is required</td><td>The operation needs the notifications write scope</td><td>Reconnect and approve the extra permission, but only if creating alerts or digests is part of your workflow</td></tr><tr><td>The connector is present but does nothing in a chat</td><td>It is not enabled for that conversation</td><td>In Claude, enable SiteTrax.io from the plus button in the message box for that conversation</td></tr></tbody></table>

## Confirming a connection is healthy

Two checks settle most uncertainty.

1. Ask the assistant to list the projects you have access to, and compare against **Projects** in the application.
2. Ask for the most recent observation of an identifier you can find yourself in the assets gallery, and compare the time and project.

If both match, the connection and its scope are correct, and any remaining difference is a question about how the data is being interpreted rather than about the connection.

<p class="callout warning">Never paste SiteTrax.io passwords, OAuth tokens or client secrets into a prompt, a configuration field or a support email.</p>

## Getting help

When reporting a connection problem, include the assistant you are using, the approximate time, the SiteTrax.io account email, the project involved, what you attempted, and any error message shown.