1
Add
Add the server
Paste one address. The client registers itself.
- Server URL
- adrails.ai/api/mcp
- Transport
- Streamable HTTP
- Sign-in
- OAuth
Setup guide
How to connect Meta Ads and Google Ads to Claude and ChatGPT
Claude and ChatGPT reach your ad accounts through an MCP connector. With Adrails, every client uses one server address and one OAuth sign-in: add the server, choose a workspace and its permissions, ask a first question, then approve each change in the browser.
Reviewed Oct 3, 2026
Per client
Where the server goes in each client
Every client uses the same server address, https://adrails.ai/api/mcp. Only the place you paste it changes.
| Client | Where you add Adrails | How you sign in |
|---|---|---|
| Claude | Settings, Connectors, Add custom connector | Connect, then allow access in Adrails |
| Claude Code | claude mcp add --transport http adrails <url> | Run /mcp, select adrails, authenticate |
| ChatGPT | Developer mode, Plugins, add a custom MCP server | OAuth, then allow access in Adrails |
| Codex | codex mcp add adrails --url <url> | codex mcp login adrails |
From the Adrails MCP documentation. ChatGPT needs Developer mode, which depends on your plan and your workspace's policy.
Product workflow
Four steps from your AI client to a first approved change
The same steps in Claude, Claude Code, ChatGPT and Codex: add the server, consent in Adrails, check the connection, then approve in the browser.
2
Consent
Choose what it can do
One workspace and the permissions this client gets.
- Read workspace data
- Required
- Stay connected
- Selected
- Prepare changes
- Owner or admin
3
Check
Ask a first question
A read confirms the connection and changes nothing.
- Tool
- integrations_status
- Returns
- Connected sources
- Listed in
- Your connections
4
Approval
Approve in Adrails
A change arrives as a link to review in the browser.
- Status
- Awaiting approval
- Approved by
- You, signed in
- Expires
- 30 minutes
Illustrative workspace data. Values show the product flow and are not customer results.
01
Before you start
You need an Adrails workspace with Meta Ads or Google Ads connected, because the assistant works on what the workspace holds. Any member can connect a client with read access. Preparing advertising changes needs an owner or admin.
Every client uses the same server address, https://adrails.ai/api/mcp, and the same sign-in. The client registers itself with Adrails, so there is no client ID, secret or API key to copy. Google Ads accounts work the same way once connected; the Google Ads platform page says whether new connections are open.
02
Connect Claude
On claude.ai and in the Claude desktop app, open Settings, then Connectors, and choose Add custom connector. Name it Adrails and paste the server URL. Leave the optional OAuth client ID and client secret empty: Claude registers itself.
Add the connector, then connect it. Adrails opens: sign in, choose the workspace and the permissions, and choose Allow access. Enable Adrails in a conversation and ask your first question. On a Claude Team or Enterprise plan, an owner of your Claude organization may have to add the connector first.
- Settings, Connectors, Add custom connector
- Name: Adrails
- Server URL: https://adrails.ai/api/mcp
- OAuth client ID and secret: leave empty
03
Connect Claude Code
From your project directory, run claude mcp add --transport http adrails https://adrails.ai/api/mcp.
Then, in Claude Code, run /mcp, select adrails and authenticate. Your browser opens Adrails for the sign-in and the consent screen. Return to Claude Code once you have allowed access.
04
Connect ChatGPT
In ChatGPT Settings, turn on Developer mode. Whether you can depends on your ChatGPT plan and your workspace's policy.
Open Plugins and add a custom MCP server named Adrails, with the server URL and OAuth as its authentication. Sign in to Adrails, choose the workspace and the permissions, and choose Allow access. Review the tools ChatGPT discovered, then add Adrails to a new conversation.
05
Connect Codex or another client
In Codex, run codex mcp add adrails --url https://adrails.ai/api/mcp, then codex mcp login adrails with the scopes you want the consent screen to offer. Start Codex and ask it to use the Adrails tools.
Another client connects when it supports remote MCP over Streamable HTTP, OAuth with dynamic client registration, and the authorization code with PKCE. Add a remote MCP server named Adrails with the server URL, choose Streamable HTTP and OAuth if it asks, and follow its sign-in link.
06
Choose the permissions
The consent screen lists only what your client asked for. Read workspace data is required. Stay connected renews access without a new sign-in until you revoke it. The write permissions start unselected: select only what you want the assistant to do.
A permission never goes beyond your role in the workspace, and each connection works in the one workspace you chose. To work in another, connect again and choose it.
- Read workspace data: required, any member
- Stay connected: selected at first, any member
- Prepare advertising changes: owner or admin
- Manage integrations, automations or campaign assets: owner or admin
- Edit Knowledge: any member
07
Check the connection and approve a first change
Ask which integrations are connected to your workspace, then ask to compare campaign performance this week with last week. Your client now appears under Your connections on the MCP page in Adrails.
Ask for a daily budget increase on your best campaign. With Prepare advertising changes, the assistant returns a proposal and its approval link. Open it, check the change field by field, and choose Approve this change or Decline. Nothing changes before you approve, and a proposal expires after 30 minutes.
08
When it does not work
If the client says it needs authentication, the connection was revoked, or it expired without Stay connected: connect again. If a write tool is missing, the connection lacks its permission or the tool needs an owner or admin: connect again and select the permission.
If the client sent two calls at once, the second is refused with Another call is running on this connection; send it again. To cut a client off, open the MCP page in Adrails and choose Revoke access under Your connections.
FAQ
Common questions
What are Claude connectors?
Claude's name for the remote MCP servers you add under Settings, Connectors. A custom connector points Claude at a server such as Adrails, and Claude calls its tools in your conversations.
Does ChatGPT support MCP connectors?
Yes, through Developer mode, which adds custom MCP servers with OAuth. Whether you can turn it on depends on your ChatGPT plan and your workspace's policy.
Is the permission prompt in Claude or Codex an approval?
No. When your client asks you to allow a tool call, that is the client's own setting. An advertising change still waits for your approval in Adrails.
How do I disconnect an AI client?
Open MCP in Adrails, go to Your connections and choose Revoke access. Removing Adrails from your client does not revoke the connection in Adrails, so revoke it there too.
Which permission lets a client prepare changes?
Prepare advertising changes, which only an owner or admin can grant. A proposal also needs an active plan.
Next