Hey I'd like to connect Claude to Mixpanel with Mixpanel MCP server. During authorisation, it mentions "Read access to user data that you have tracked to Mixpanel. Grants the ability to run queries and export data" Is there a way to exclude user/profile info access? is it possible to share only ids and event?
That's a great privacy-conscious question. Based on the available documentation, here's what I can tell you:
Connecting Claude to Mixpanel MCP
Connecting Claude is straightforward — you go to claude.ai/customize/connectors, search for Mixpanel, click Connect, and complete the OAuth flow. [Connecting to MCP]
Limiting Access to Only IDs and Events (Excluding Profile Data)
The knowledge sources do not describe a way to selectively exclude user/profile properties from the MCP OAuth scope at the authorization step. The required OAuth scopes (projects, analysis, events, insights, segmentation, retention, data:read, funnels, flows, data_definitions) are mandatory and don't offer a granular "no profile data" option. [Building Custom Integrations]
However, there are indirect ways to limit what the AI assistant can actually see:
Data Classification: Mark sensitive user/profile properties (like $email, $phone, etc.) as "Classified" in your project. Classified properties are only visible to users explicitly granted access to them. If the Mixpanel user account used for the MCP OAuth connection doesn't have classified data access, those properties won't be exposed. [Data Views & Classification]
Data Views: Create a restricted Data View that excludes PII-containing events or properties, and ensure the connecting user is scoped to that Data View. The MCP server respects existing project-level permissions — users can only access data from projects and views they already have permission to see. [MCP Permissions]
Dedicated MCP User Account: Consider creating a Mixpanel user account with minimal permissions (e.g., scoped to a restricted Data View with no classified data access) specifically for MCP connections.
Important Caveat
The docs also note that your Mixpanel data is sent to the AI provider (Claude/Anthropic in this case), so you should review Anthropic's data handling policies before connecting projects that contain personal data. [Security Considerations]
In summary, you can't exclude profile data at the OAuth scope level, but you can use Data Views and Data Classification to effectively limit what's visible to the connecting account.
ℹ️ Tip: Mention @kapa.ai in the thread for follow-ups or DM the app for 1-to-1 help.
That's a great security-conscious question. Based on the available documentation, there is no built-in way to granularly exclude user/profile data from the MCP OAuth scope during the authorization flow. The required scopes are defined as a fixed set:
Mandatory scopes: projects analysis events insights segmentation retention data:read funnels flows data_definitions
These scopes are requested as a bundle, and the documentation does not describe a way to strip out profile/user data access specifically from the Claude connector flow.
However, there are two Mixpanel-native approaches that can limit what user/profile data is exposed through the MCP connection:
1. Data Views
You can create a Data View that excludes certain events or filters out specific user profile properties, and then ensure the connecting user only has access to that restricted Data View. Since MCP respects existing project-level permissions, the AI assistant will only see what that user can see. [Data Views & Classification]
2. Classified Data
You can mark sensitive user/profile properties as Classified in Lexicon. Users without "Can View" permission for classified data won't be able to use those properties in queries — this restriction carries over to MCP sessions as well, since all existing roles and project-level permissions remain in effect. [Data Views & Classification]
Important caveat
The MCP documentation explicitly warns:
"Your Mixpanel data is also sent to whichever AI provider you're using (Claude, ChatGPT, etc.), so review that provider's data handling policies before connecting." [MCP Security]
So the most practical path to limiting profile data exposure is to use Data Views + Classified Data to restrict what the connecting Mixpanel user can access, rather than trying to scope down the OAuth grant itself.
ℹ️ Tip: Mention @kapa.ai in the thread for follow-ups or DM the app for 1-to-1 help.
That's a great privacy-conscious question. Based on the available knowledge sources, here's what I can tell you:
Connecting Claude to Mixpanel MCP
Connecting Claude is straightforward — you go to claude.ai/customize/connectors, search for Mixpanel, click Connect, and complete the OAuth flow. [MCP connecting]
Limiting Access to User/Profile Data
The knowledge sources do not describe a way to exclude user/profile properties from the OAuth scope at the MCP connection level. The required OAuth scopes are fixed and include broad read access (projects, analysis, events, insights, etc.). [MCP OAuth]
However, there are Mixpanel-side controls you can use to limit what the AI assistant can actually see:
1. Data Classification (Enterprise only)
You can mark sensitive user/profile properties as Classified in Lexicon. Users (and by extension, the MCP connection acting on their behalf) without "Can View" permission on classified data won't be able to use those properties in queries. [Data Classification]
2. Data Views
You can create a Data View that excludes specific events or filters out sensitive user profile properties, and ensure the connecting user only has access to that restricted view. This effectively limits what the MCP session can query. [Data Views]
3. General Access Controls
The MCP server respects existing Mixpanel roles and project-level permissions — users can only access projects and data they already have permission to view. [MCP security]
Important Caveat
The docs also note that your Mixpanel data is sent to the AI provider (Claude/Anthropic in your case), so you should review Anthropic's data handling policies before connecting projects that contain personal data. [MCP security]
In summary: you can't surgically exclude user profile data at the OAuth scope level, but you can use Data Views and Data Classification within Mixpanel to control what's visible to the connecting user account.
ℹ️ Tip: Mention @kapa.ai in the thread for follow-ups or DM the app for 1-to-1 help.
Based on the available documentation, there is no built-in way to granularly exclude user/profile data from the MCP OAuth scope during the authorization flow. The mandatory scopes are defined as a fixed bundle and cannot be selectively removed. [MCP OAuth docs]
However, there are two Mixpanel-native approaches that can limit what user/profile data is exposed through the MCP connection, since MCP respects all existing project-level permissions and roles:
1. Data Views
Create a Data View that filters out specific user profile properties or excludes certain events, then ensure the connecting user is assigned only to that restricted Data View. The AI assistant will only see what that user can access. [Data Views & Classification]
2. Classified Data
Mark sensitive user/profile properties as Classified in Lexicon. Users without "Can View" permission for classified data cannot use those properties in queries or filters — this restriction applies to MCP sessions as well. [Data Views & Classification]
Important caveat
The MCP documentation explicitly warns:
"Your Mixpanel data is also sent to whichever AI provider you're using (Claude, ChatGPT, etc.), so review that provider's data handling policies before connecting." [MCP Security]
So the most practical path to limiting profile data exposure is combining Data Views + Classified Data to restrict what the connecting Mixpanel user can access, rather than scoping down the OAuth grant itself.
ℹ️ Tip: Mention @kapa.ai in the thread for follow-ups or DM the app for 1-to-1 help.
