Connect Your Own AI (MCP)
Use your AIdata properties from claude.ai, Claude Desktop, Claude Code, Cursor or any MCP-compatible AI client with one personal token.
External AI access lets you use your AIdata properties from your own AI assistant - claude.ai, Claude Desktop, or any client that supports MCP (Model Context Protocol). Your assistant gets the same data tools as the built-in chat, and usage is covered by the AI subscription you already pay for. No API key involved.
Requirements
- Any plan. Free includes 300 tool calls a month per team; Pro and Business are unlimited.
- External AI Access is switched on in Team Settings. It is on by default; a team owner can switch it off.
- For claude.ai, custom connectors require a paid Claude plan. Claude Desktop and other MCP clients work with whatever plan they normally need.
Setting It Up
1. Check external access is on (team owner)
External AI Access in Team Settings is on by default, so there is usually nothing to do. It is the master switch for the whole team: while it’s off, no member’s connector can reach that team’s properties.
2. Get your personal connection details
Go to Settings → Connected AI and click Show my connection details. You get three values:
- Server URL - the AIdata MCP endpoint,
https://app.aidata.stream/mcp/u. It is the same for everyone and contains no secret. - Authorization header -
Bearerfollowed by your personal token. This is what identifies you. - Connection URL, under Clients that only accept a URL - the same endpoint with your token built into the address, for clients that have nowhere to enter a header.
Your token is personal: it identifies you, covers every property you can access across all your teams, and enforces your permissions on each property. Each teammate gets their own token - never share yours.
Treat the token and the URL like a password. Anyone who has either can read every property you can access and write annotations in your name. If one ever leaks, click Regenerate token - every copy of the token and the URL stops working immediately, and your teammates are unaffected.
3. Add it to your AI client
Use the header form wherever your client supports request headers. The token then never appears in a URL, so it stays out of server logs, proxies and browser history.
claude.ai and Claude Desktop: Settings → Connectors → Add custom connector. Give it a name (e.g. “AIdata”) and paste the Server URL. Under Authentication choose No sign-in - claude.ai may mark Sign in now as “Detected”, but that is only its guess for any server that asks for credentials; this connector uses a token, not OAuth. Choose it first: the OAuth client section disappears and Request headers becomes the credentials section. Click Add header, pick authorization as the name and paste the Authorization header value from the app (Bearer …) as the value, leave Required ticked, then save.
Claude Code:
claude mcp add --transport http aidata https://app.aidata.stream/mcp/u \
--header "Authorization: Bearer <your token>"
Cursor and other MCP clients: add a remote (HTTP) MCP server with the Server URL and a request header Authorization: Bearer <your token>.
Clients that only accept a URL: paste the Connection URL as the server URL instead. No header is needed - the URL itself grants access - but the token travels in the address, so treat every place you paste it as a copy of your password.
4. Use it
In a chat, make sure the connector is enabled in the tools menu, then just ask. Good first prompt: “List my AIdata properties.” The AI will discover your properties, see which data sources each has connected, and pick the right one for follow-up questions.
Signing In Instead of Using a Token
Clients that support OAuth can connect without copying a token. Use the Server URL https://app.aidata.stream/mcp (without /u) and sign in with your AIdata account when the client asks. AIdata shows a consent page, Authorize access to AIdata, naming the application and what it will be able to do. Click Allow access and you are sent back to the client. The connection then has exactly the access your token would have: the same properties, the same permissions, the same write rules.
- claude.ai and Claude Desktop: add the custom connector with this URL and choose Sign in now.
- Claude Code:
claude mcp add --transport http aidata https://app.aidata.stream/mcp, then run/mcpin a session and choose Authenticate for aidata. Your browser opens the consent page. - Access ends by itself after 30 days without use; sign in again from the client to continue. To stop a client sooner, remove the connector in that client.
- Your personal token keeps working alongside. The two are independent: regenerating the token does not disconnect a signed-in client.
Only redirects back to claude.ai or to a client running on your own computer (such as Claude Code) are accepted. For any other client, use the token method above.
What Your AI Can Do
- Read everything the built-in chat can: traffic reports, search rankings, ad performance, CRM pipelines, Core Web Vitals, Cloudflare edge traffic and AI crawler data, sitemap change history, annotations, and the knowledge base of every connected property.
- Run your skills: skills appear in your client’s prompts menu, and the AI can also list and fetch them itself - ask for a skill by name and it follows your playbook exactly.
- Write annotations: on properties where you have write permission, the AI can create and update custom annotations, attributed to you and visible to your team in the app.
- Save team memories: on properties where memory is enabled and you have write permission.
- Update the to-do list: on properties where you have write permission, the AI can add follow-ups to the to-do list and tick them off, the same as in the built-in chat.
Everything else is read-only. The connector can never change anything in your connected platforms - it can’t touch Google Analytics settings, edit campaigns, or modify your WordPress site.
Permissions and Security
- Your permissions travel with you. The connector only sees properties you can access, and property-level permissions (read/write/admin) apply to every call - the AI is told your permission level per property.
- Team-level kill switch. A team owner turning External AI Access off instantly removes that team’s properties from every member’s connector, without invalidating anyone’s URL. Re-enabling restores access.
- Personal rotation. Regenerating your token permanently invalidates your old token and Connection URL everywhere you used them. Full lockdown for a team = every member rotates, owner disables.
- Leaving a team removes its properties from your connector automatically.
Notes on Metered Tools
Some tools spend money when called - for example DataForSEO backlink lookups. These bill the team that owns the property the AI passes, and the AI is instructed to tell you which property’s connection it is using when it looks up a domain that isn’t that property’s own site. Connection admins can restrict lookups to specific domains in the DataForSEO and PageSpeed connection settings (allowed domains).
When the Built-In Chat Is Better
The connector is ideal for quick questions from wherever you already work. The built-in chat is still sharper for heavy analysis - it stores large results in a per-conversation database and runs real SQL over them - and reports, report history, and sharing all live in the app. See the announcement post for an honest comparison.
Troubleshooting
- “External AI access is switched off for all of your teams” - every team you belong to has External AI Access switched off, or none of them has a property you can access. Ask a team owner to switch it on in Team Settings.
- “…has used all 300 external AI tool calls included in its plan this month” - the team is on the Free plan and has reached its monthly allowance. It resets on the 1st of the month (UTC); Pro and Business have no limit. Listing properties, skills and these docs still works. The owner sees this month’s usage under Team Settings → External AI Access.
- A property is missing - either its team has external access disabled, or you don’t have access to that property in the app. The connector never shows more than the app does.
- “Unauthorized MCP access”, or the client says a login is required - the Server URL was added without the Authorization header, or the header value is wrong. It must be
Bearerfollowed by your full token, exactly as shown under Settings → Connected AI. If you regenerated your token, update the header everywhere you added it. - The AI doesn’t use the connector - check it’s enabled in the chat’s tools menu, and ask explicitly (“use AIdata to…”) the first time.
- Writes are refused - annotation, memory and to-do writes need write permission on that property; ask a team admin to grant it.
- Sign-in fails with “invalid_client” - the client sent a redirect address AIdata does not allow. Only claude.ai and clients running on your own computer (such as Claude Code) can sign in; use the token method for anything else.