xMatix publishes one endpoint on the open Model Context Protocol (MCP). An assistant that speaks it — Claude on the web or desktop, ChatGPT, Claude Code and similar hosts — can connect to your organization, sign you in with your normal xMatix credentials and then ask your records, run your reports and dashboards, search your organization's knowledge and, where your administrator allows it, create and update records. Every call runs as you, inside your permissions, and is audited. This page is the user's and administrator's view; the protocol, scopes and full tool list are on the MCP endpoint page for developers.
Before you can connect
Three things must be true, and only the first is yours to change:
- You hold a Sense AI seat in the organization — see Licensing. The connector runs the same Sense tool surface as the in-product assistant and follows the same seat.
- An administrator has switched on Allow external assistants for the organization. It is off by default for every organization.
- If the administrator keeps an Approved assistants list, the assistant you are using is on it.
If the assistant reports that it cannot find an authorization server, or your organization is not offered after sign-in, one of these is the reason. Ask your administrator rather than retrying.
Connecting an assistant
Your administrator gives you the organization's MCP endpoint URL. The steps differ slightly per assistant, but each follows the same shape: paste the endpoint, sign in with your xMatix credentials in the window that opens, choose the organization if you belong to more than one, and approve the requested access.
- Claude (web or desktop): Customize → Connectors → Add custom connector, paste the endpoint URL and follow the sign-in prompt.
- ChatGPT: Settings → Connectors → Create, paste the same URL. The connector also works in deep-research mode.
- Claude Code: add the endpoint as an HTTP MCP server from the terminal; the sign-in opens in your browser and returns to the terminal when done.
The consent screen you see is rendered by xMatix. It names the assistant, the organization the connection is bound to, and the access that will actually be granted after your organization's ceiling is applied — if your administrator has not allowed writes, no write access is offered, whatever the assistant asked for.
What the assistant can do as you
| The assistant can | The assistant cannot |
|---|---|
| Discover your organization's entities and fields by name | See a record or field your own permissions do not allow |
| Query, filter, count and aggregate records — aggregations come back as charts where a chart fits | Change data unless the organization has allowed writes, and then only within your permissions |
| Run the saved reports and dashboards you can already open | Use a tool that a Sense agent could not be granted in AI Studio |
| Search the knowledge collections that ground your Sense agents, and read the platform documentation | Keep working after an administrator disconnects it or switches the feature off |
| Read and change the organization's setup — data model, layouts, business rules, scripts, workflows, integrations — if you are an administrator and writes are allowed | Act for anyone but the signed-in user |
| Reason with its own model about everything it reads | Hand work to Sense or the Setup Copilot — your organization's AI credits are never spent by an external assistant |
A good first question is to ask the assistant who it is acting as. It answers with your user, your organization and the effective access. Then ask something you know the answer to, and finally something you are not allowed to see — the assistant should report that record as not accessible, because the query ran as you.
Who is responsible for what
Nothing about the connector widens access. Each of three layers can only narrow the one before it: the access granted at sign-in, the organization's Sense capability grants, and your own security profiles, record security and restriction rules at the moment a call runs. Every call is written to the audit trail as you; setup changes appear in the setup audit trail exactly as if you had made them in the Setup screens.
What the assistant provider does with the content of a conversation is governed by your organization's agreement with that provider, exactly as it is when a colleague pastes an export into it today. The connector changes how the data arrives — permission-checked, cited and audited — not who you are talking to.
For administrators
External assistants are governed from Setup → Sense AI Studio → External Assistants. The page shows the organization's MCP endpoint URL, counters for assistants, writes and active connections, the list of every connection — assistant, user, access, connected and last-used times, status — with a per-row Disconnect, and a Settings dialog with three controls:
| Control | Default | What it decides |
|---|---|---|
| Allow external assistants | Off | Whether any assistant may act as this organization's users. Switching it off ends every session immediately. |
| Allow writes (create/update records; setup changes for administrators) | Off | Whether write access can be granted at all. Off caps every connection at read-only, whatever the assistant requests. On also unlocks the setup tools for administrators. |
| Approved assistants | Any | An optional allow-list of assistant clients. Anything not listed is refused at sign-in. |
Policy is re-evaluated on every token refresh and on every call, so changing a setting applies to existing connections without anyone reconnecting. Most organizations run read-only for the first weeks and switch writes on once the team trusts the queue. The administrator's rollout guide, with the checks to run before announcing the connector, is Connecting Claude or ChatGPT to xMatix.
Common questions
Do I need to install anything?
No. The endpoint is part of the platform and every supported assistant has a built-in way to add an MCP connector. You need the endpoint URL from your administrator and your normal xMatix sign-in.
Why does the assistant see fewer tools than the Sense chat in xMatix?
Two reasons, both intended. The access granted at sign-in may exclude writes because your organization has not allowed them, and tools that need the browser — such as acting on the screen — are only offered to surfaces that have one. Everything the assistant does see runs through the same permissions as the in-product chat.
Can I connect more than one organization?
Each connection is bound to one organization, chosen at sign-in. To work with a second organization, connect again and choose it; the assistant then holds one connection per organization.
How do I disconnect?
Remove the connector from the assistant's settings, or ask your administrator to disconnect it from the External Assistants page. Either ends the connection immediately.
Related topics
- The MCP endpoint — protocol, scopes and the full tool list
- Asking Sense about your data
- How Sense AI is governed
- Sense in Claude and ChatGPT — the product overview