Connect AI to your tools (MCP connectors)
A connector links the Support Station AI to one of your other systems, for example your billing system. The AI can then read facts from that system when it writes a reply draft for your agents.
What Is a Connector?
MCP (Model Context Protocol) is an open standard. It lets an AI use the tools of an outside server. Many products, for example Stripe, give you an MCP server.
A connector is one MCP server that your organization adds to Support Station. Each server has a list of tools. A tool does one task, for example "find a customer" or "list invoices".
Example: A customer asks why a payment failed. The AI uses your billing connector to read the customer's invoices. It then writes a draft that gives the reason. Your agent reads the draft, edits it, and sends it.
Connectors work only in AI reply drafts. The AI chat on your help center and in the widget does not use them.
Plan Requirements
Connectors are available on all paid plans.
| Plan | AI Connectors |
|---|---|
| Free | ✗ |
| Starter | ✓ |
| Standard | ✓ |
| Pro | ✓ |
| Enterprise | ✓ |
Only organization owners and admins can add, change, or remove connectors.
On the Free plan, the AI Connectors settings show the button Upgrade your plan. It opens your billing settings.
If your organization moves to the Free plan, you can still see and remove your connectors. You cannot add or change them.
Adding a Connector
Step 1: Open the Connector Settings
- Go to Settings > AI Connectors
- Click Add connector
Step 2: Enter the Connector Details
| Field | Description |
|---|---|
| Name | A name for your team, for example "Stripe". Each connector needs a different name. |
| URL | The address of the MCP server. It must start with https://. |
| Secret token | Optional. Support Station sends it to the server in each request. Get it from the system that you connect. |
Step 3: Save
When you save, Support Station connects to the server and reads its list of tools. If the connection fails, the connector is not saved. You see the error, and you can correct the details.
Tool Types
The settings show the tools of each connector in two groups.
| Group | What the AI Can Do | Default |
|---|---|---|
| Research tools | The AI uses the tool alone to look up facts. The tool only reads data. | On, if your organization has fewer than 30 tools on |
| Action tools | The AI can only suggest the action. An agent must click a button to run it. | Off |
The server tells Support Station which tools are read-only. A tool that the server does not mark as read-only is always an action tool. If the server says that a tool can delete data, the tool shows the label can delete data.
You can turn each tool on or off. You can also move a research tool to the action tools, and back again. You cannot make a tool a research tool if the server does not mark it read-only.
When you turn on an action tool, the AI can suggest that action on a ticket. The action runs only when an agent clicks the button. For more information, see AI reply drafts.
Limits
| Limit | Value |
|---|---|
| Connectors for each organization | 5 |
| Tools that are on, for all connectors together | 30 |
When you add a connector, Support Station turns on its research tools until your organization has 30 tools on. The other tools stay off. You can turn off a tool to make space for a different tool.
Support Station also limits what it reads from a server:
- It imports at most 200 tools from one connector. It does not import the tools after the first 200.
- It does not import a tool that has a very long name or a very long input schema.
- It shortens a very long tool description.
- If two tools have the same name, it imports only the first one.
- If the tool list of a server is very long (more than 5 pages), the connection fails.
Refreshing the Tool List
The tool list of a server can change. Click Refresh to read the list again.
When you refresh:
- New tools get the default settings in the table above. A new research tool is on only if your organization has fewer than 30 tools on.
- Tools that the server removed are removed from Support Station.
- A tool that is no longer read-only moves to the action tools and is turned off. An admin must turn it on again.
- All other tools keep their settings.
Testing a Connector
Click Test on a connector to connect to the server again. The Test button is always available, also when the connector works. If the connection fails, the settings show the last error.
Changing or Removing a Connector
You can change the name, the URL, and the secret token. When you change the URL or the token, Support Station connects to the server again before it saves.
When you change the URL of a connector that has a secret token, you must enter the token again or click Remove token. Support Station does not send a saved token to a new address. Until you do one of these, you cannot save. You can also turn off a whole connector. The AI then does not use any of its tools.
When you remove a connector, its tools are removed too.
Security
The Secret Token
Support Station keeps the secret token encrypted. After you save it, Support Station never shows it again. The settings show only if a token is set. To change the token, enter a new one.
The token is never sent to your browser or to the AI provider.
HTTPS Only
The connector URL must use HTTPS. Support Station refuses addresses inside private networks.
OAuth Sign-In
Connectors do not support OAuth sign-in (a sign-in through another service). A connector uses a secret token or no token. If a server accepts only OAuth sign-in, you cannot connect it at this time.
Troubleshooting
The Connector Does Not Save
Check:
- The URL starts with
https://and is the address of an MCP server - The secret token is correct
- The server is online
A Tool Is Not Used
Check:
- The tool is on
- The connector is on
- The 30-tool limit is not the cause. A research tool stays off if your organization had 30 tools on when the tool was added. Turn off a different tool, then turn this tool on.
- The tool was imported. A tool with a very long name or schema, or a tool after the first 200, is not imported.
- Your organization has enough AI credits. A draft needs 6 available credits to use connectors. See Understanding AI Credits.
