Use it in Claude and ChatGPT

Your gateway gives Claude and ChatGPT two connectors. You add each one as a custom connector, with your gateway’s address followed by the connector’s path. Your cloud guide’s last step lists both addresses.

Connector	Address	Who adds it	What it does
Data	https://<your gateway>/mcp	Everyone who asks questions about company data	Lists the data you may use and answers with read-only queries
Admin	https://<your gateway>/admin/mcp	Portward admins	Connects sources, sets what the AI and people see, and shows answer records

Claude

Which plans can add it

Anthropic offers custom connectors on the Free, Pro, Max, Team and Enterprise plans. On Free you can add one custom connector, so only the data connector or the admin connector. On Team and Enterprise, only some people can add one for the organization, as the section below explains. Anthropic’s custom connectors guide has the current rules for Claude’s plans.

Add it yourself on Free, Pro or Max

1. In Claude, open Customize, then Connectors.
2. Select + Add, then Add custom connector.
3. Enter a name people will recognize, such as “Portward” or “Portward admin”, and the connector’s address. Select Continue.
4. Claude shows the sign-in settings it found on your gateway. Leave Authentication as Claude detected it, Sign in now. Under OAuth client, make sure Register automatically is chosen. Your gateway accepts only apps that register with it, so Claude has to register itself. Select Add.
5. Sign in with your work account when Claude asks. If it doesn’t ask, select Connect next to the connector. If the connector shows Disconnect and This connector has no tools available instead, Claude kept an old sign-in for that address. That can happen with a connector you just added, when the address was added before. Select Disconnect, reload the page, then select Connect.

Add it for your whole organization on Team or Enterprise

On Claude Team, only Owners and Primary Owners can add a custom connector. On Claude Enterprise, people with a custom role that includes Manage access to Libraries can too (Anthropic’s guide). If you manage Portward but aren’t one of them, ask an owner to do these steps.

1. Open Organization settings, then Connectors, and select Add.
2. Point to Custom and choose Web.
3. Enter the name and the connector’s address, and select Continue. Under OAuth client, make sure Register automatically is chosen, then select Add.
4. Each person then opens Customize, then Connectors, finds the connector (it has a Custom label) and selects Connect. They sign in with their own work account.

Adding a connector for the organization doesn’t give anyone access to data. The gateway checks each person at sign-in against the users and groups you set for that connector. Add the data connector this way for everyone. If your Portward admins can’t add connectors themselves, the owner adds the admin connector too. People who aren’t Portward admins see Not allowed when they connect it.

Claude Team and Enterprise owners can also turn off the tools that show tables and forms in the chat, in Organization settings, then Connectors (Anthropic’s guide). Leave query on for the data connector and setup on for the admin connector. Without them, people see only the AI’s text, with tokens in place of real values, and admins can’t connect sources.

Turn it on in a chat

Select + at the bottom left of the message box, then Connectors, and switch the connector on for that conversation.

ChatGPT

Which plans can add it

OpenAI calls a connector like Portward’s a custom MCP server. It offers them on ChatGPT Business, Enterprise and Edu, in ChatGPT on the web. They don’t work in the mobile apps. On Business, only workspace admins and owners can add and publish one. On Enterprise and Edu, admins can also let chosen members add them. OpenAI says Pro accounts can connect custom MCP servers for reading and fetching only. These steps are for Business, Enterprise and Edu. OpenAI’s custom MCP servers guide has the current rules for ChatGPT’s plans.

Add it

OpenAI is moving these screens from Apps to Plugins. These steps follow OpenAI’s guide as it read on 2026-10-27.

1. In ChatGPT on the web, open Plugins, select the plus button, then Add custom MCP server.
2. Enter a name, such as “Portward”. Under Connection, enter the connector’s address.
3. Choose OAuth for authentication and leave the client ID empty. ChatGPT then registers itself with your gateway.
4. Read the risk warning, tick I understand and want to continue, and select Create as a plugin.
5. Sign in with your work account when ChatGPT asks. ChatGPT then lists the tools it found on your gateway.

Publish it for your workspace

Only workspace admins and owners can publish (OpenAI’s guide).

1. In ChatGPT’s Admin Console, choose your workspace, open Plugins, then Manage Legacy Apps.
2. Open Drafts, select Publish next to the connector, and read the warnings. On Enterprise and Edu, Configure Access chooses which groups get it.
3. People find it under Settings, then Plugins, and sign in with their own work account.

ChatGPT keeps the list of tools it saw when the connector was published. When a Portward update adds or changes a tool, calls to it can fail until an admin refreshes the tool list in the workspace’s app settings.

Use it in a chat

Type @ in the message box and choose the connector. The choice applies to that message. To get new data in a later message, choose it again.

The consent prompt

The first time you connect, the gateway sends you to your company’s sign-in page. After you sign in, the gateway shows Allow this app? with these facts:

• App: Claude or ChatGPT. The gateway names the app from the address it sends you back to. An app can’t fake that address.
• Calls itself: the name the app gave itself, shown only when it differs.
• Signed in as: your work account, by its email address when your directory sends a verified one, otherwise by your name or sign-in id.
• Sends you back to: the app’s sign-in address.

For the data connector it adds “It will see your company data through Portward, with your access. Its answers are recorded.” For the admin connector, “It will manage Portward for you, including connections and what the AI may see.”

Select Allow only if you added Portward in that app yourself a moment ago and Signed in as is you. Otherwise select Cancel. If it names someone else, sign out of your directory in that browser, then connect again. The gateway asks once for each app, person and connector.

If you see Not allowed instead, the gateway doesn’t count you among that connector’s users or admins. Ask whoever manages Portward to add you, then connect again. The gateway’s log names your sign-in id, so they can find you.

Claude and ChatGPT can stay signed in for up to 30 days. Each hour the gateway asks your directory again, so someone you disable there loses access within the hour. Google Workspace can be different. If Google doesn’t let the gateway renew the sign-in (it sends no refresh token), the gateway can’t ask Google again. The sign-in then ends after 12 hours, and someone you disable in Google keeps access until it does.

Ask a first question

With the data connector on, ask what you can use, such as “Which datasets can I ask about?”. The AI calls list_datasets, which lists the datasets, columns and definitions your groups may see. Then ask a real question, such as “How many open invoices does each customer have?”.

The AI answers with query, one read-only SQL query per call. It also writes down how it read your question, such as which status counts as open. Your admins see those assumptions and can turn them into definitions.

Data connector tool	What it does
list_datasets	The datasets, columns, definitions and links you may use
describe	One dataset in detail, with a few sample rows (sensitive values as tokens)
query	Runs one read-only query and shows the results table
suggest_links	Finds columns in another source that likely match, and shows a picker
save_link	Saves the match you picked

Admins ask the admin connector in plain words, for example “Set up a connection” (setup), “Show only totals for the salary column” (propose_access), “What needs my attention?” (governance) or “Show the evidence for the last answer” (evidence).

Tokens and real values

For each column, your admin decides what the AI sees: the real value, a token, or only totals and averages. A token looks like EMAIL_1a2b3c4d5e6f7a8b, a type followed by 16 characters. A value of one type always gets the same token, in every source, so the AI can count, group and match rows across sources by token without seeing what the tokens stand for. Setup suggests the types EMAIL, PHONE, NAME, ADDRESS, DOB, GOVID, ACCOUNT and IP, and admins can add their own.

Real values show only in the results table that opens above the AI’s answer. Your gateway sends them straight to that table in your browser. The gateway never puts them in what it sends the AI, and the table says so: “Real values are shown to you only.” Your admin also decides what people see in each tokenized column:

Setting	In the table
Real value	The value
Masked	Dots. For values longer than 8 characters, the last 4 show.
Click to show	Dots and a Show button. Each Show is recorded.
Hidden	The word Hidden

To find one person or account, type a name or email into Find in these results. The table keeps the matching rows. The AI never sees what you typed.

If your admin lets people share a column, the table has a Send real values to Claude (or ChatGPT) button. Select it, then Yes, send to Claude to send the values your admin lets people share from that answer. The rest stay tokens. ChatGPT gets them at once. Claude puts them in your message box, and you press send. Each send is recorded.

About this data, under the table, shows the definitions and context the AI was given.

Find the answer record

Every query makes an answer record, and so does every describe, for its sample rows. The results table’s footer shows “Answer record” and the record’s ID, 32 letters and digits. The AI is asked to list the records it relied on at the end of its answer. If it doesn’t, use the ID in the table’s footer.

Admins can see what’s in a record two ways:

• Ask the admin connector “Show the evidence for answer record” followed by the ID, or “for the last answer”. The evidence tool opens the evidence pack, which you can download. It holds no row values. It lists:

  • who asked and when, and the SQL
  • the tokenized columns, and the definitions and rules in force
  • the AI’s assumptions
  • a fingerprint (hash) of the result
  • later reveals and shares
  • whether every log passes its tamper check

  It doesn’t hold the AI’s reply, which stays in Claude or ChatGPT.

• Open the admin console at https://<your gateway>/console/, sign in with your work account, open Records and search for the record ID.

If something’s wrong

What you see	What to do
Claude or ChatGPT can’t connect	Open https://<your gateway>/health in a browser. If it doesn’t answer, see your cloud guide’s “If something’s wrong”.
Claude’s sign-in fails before your company’s sign-in page opens	Remove the connector and add it again with Register automatically under OAuth client.
Claude shows the connector with Disconnect and This connector has no tools available	Claude kept an old sign-in for that address, for example from before the gateway was reinstalled. Select Disconnect, reload the page, then select Connect and sign in.
Not allowed after you sign in	You aren’t in that connector’s users or admins yet. Your Portward admin adds you.
The results table says “Real values aren’t available here”	Ask the question again in the chat.
ChatGPT’s calls fail after a Portward update	A workspace admin refreshes the connector’s tool list.
Claude shows only text, no table	A Claude owner may have turned off the query tool’s interactive view. Turn it back on in Organization settings, then Connectors.

For anything else, write to support@portward.ai. Say which step you were on and what the screen said, and leave your license key out.