How to get Email Id from the Agentspace/Gemini Enterprise Session

[rghv-cwx]

I have developed a multi agent system using adk, which internally uses a couple of tools. One of the tool is using mcptoolset to connect with the bigquery data (the mcp is deployed as a separate cloud run service).
We have also deployed this in agent engine (did the testing on playground working fine) and then exposed to Gemini Enterprise UI.

Problem: I want to use the email id using which the user has logged inside the AgentSpace/Gemini Enterprise, as one of the input in the tool -which will help in filtering the data. So that by default the user is only able to see data related to the email id.

[adk-bot]

Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: To get the user's email, you should protect your tool with an OpenID Connect (OIDC) authentication scheme and request the "email" scope. The user's identity information will then be available in the tool_context.auth_response after the user consents.

---

Hello! That's a great question. Accessing user-specific information like an email address for data filtering is a common requirement for creating personalized and secure agents.

In ADK, accessing user PII such as an email is rightly handled through a secure authentication flow rather than being automatically available in the session state. While the ADK context provides a user_id for session tracking [6, 12], obtaining the user's actual email address requires an explicit request and user consent. This is a security feature to protect user privacy.

The recommended way to achieve this is by using ADK's authentication capabilities:

1. Configure an Authentication Scheme: You need to set up an OpenIdConnectWithConfig authentication scheme. This tells your agent how to perform an OIDC/OAuth2 flow to get the user's identity from an Identity Provider (like the one used by Gemini Enterprise). When you define this, you must include "email" in the list of scopes you request. [14, 17]

   from google.adk.auth.auth_schemes import OpenIdConnectWithConfig
   
   # This configuration tells ADK how to request user identity.
   # The actual endpoints will be specific to the environment.
   auth_scheme = OpenIdConnectWithConfig(
       authorization_endpoint="<your_idp_auth_endpoint>",
       token_endpoint="<your_idp_token_endpoint>",
       scopes=['openid', 'profile', "email"] # Request the email scope!
   )

2. Protect Your Tool: Apply this authentication scheme to the tool that needs the email address. When this tool is called for the first time, the ADK framework will automatically trigger the authentication flow, prompting the user for consent to share their profile information, including their email.

3. Access Credentials in Your Tool: Once the user authenticates, the resulting credentials (typically an ID token) will be available inside your tool via the ToolContext. You can access it from the auth_response attribute. Your tool logic would then need to decode the ID token (which is a JWT) to extract the email claim. [4]

   from google.adk.tools import ToolContext
   import jwt # You might need a library like PyJWT to decode the token
   
   def my_bigquery_tool(tool_context: ToolContext, query: str):
       """
       A tool that filters BigQuery results based on the user's email.
       """
       user_email = None
       id_token = tool_context.auth_response.get("id_token") # Path may vary
   
       if id_token:
           # The ID token is a JWT containing user claims.
           # Decoding it reveals the information you requested via scopes.
           # Set verify=False for this example, but in production, you should
           # properly verify the token signature against the IDP's public keys.
           decoded_token = jwt.decode(id_token, options={"verify_signature": False})
           user_email = decoded_token.get("email")
   
       if user_email:
           # Now you have the user's email to use in your BigQuery query
           print(f"Filtering data for user: {user_email}")
           # ... add email to your BigQuery filtering logic ...
       else:
           # Handle cases where the email could not be retrieved
           raise Exception("Could not retrieve user email for data filtering.")
   
       # ... rest of your tool logic ...

This approach ensures that your agent securely and transparently requests access to user data, aligning with standard security and privacy best practices. [18]

---

References:

[1] Tool Context: https://github.com/google/adk-python/blob/main/docs/tools-custom/index.md#tool-context [4]

[2] Authentication Docs: https://github.com/google/adk-python/blob/main/docs/tools/authentication.md [17]

[3] Auth Example Snippet: https://github.com/google/adk-python/blob/main/examples/snippets/tools/auth/tools_and_agent.py [14]

[jpassing]

When you register an ADK agent in Gemini Enterprise and configure authorization, Gemini Enterprise forwards the user's access token to the ADK agent, and the ADK makes it available in the session state. No OpenIdConnectWithConfig required.

Currently, the ADK doesn't seem to have built-in support for injecting the token as an AuthCredential into the tool call, but it's not too difficult to add that functionality yourself.

[AniGunner]

@jpassing Is there any way for an ADK tool to access the signed-in user’s email address from Gemini Enterprise when OAuth is not enabled for the agent?

In our setup, the user logs in to Gemini Enterprise using Workforce Identity Federation with Microsoft Entra ID. The ADK agent itself currently does not have authentication/OAuth enabled, but we need to capture the logged-in user’s email address inside the ADK tool, mainly to pass it to the backend system as the requester.

From what I understand, if the ADK agent is registered with Gemini Enterprise / Agent Platform, Gemini Enterprise may pass the user email to the agent for personalized interactions. However, for tool-level delegated access or confirmed identity claims, we may still need an OIDC/OAuth-based auth scheme with the email scope.

Can you please confirm whether the user email is available directly in the ADK runtime/tool context without enabling OAuth, and if yes, which field/context object should we read from?

[jpassing]

If you register an agent in Gemini Enterprise, authentication is entirely controlled by Gemini Enterprise -- so there's usually litltle reason to set up OAuth in the agent itself at all.

By default, Gemini Enterprise authenticates to Agent Engine using its own service identity and doesn't pass any information about the end user to the agent. However, you can change that by configuring a tool authroization for the respective agent in Gemini Enterprise.

• To have Gemini Enterprise pass an Entra access token, you can do the following

  1. Create an app registration in Entra
  2. Configure a tool authorization that uses this app registraration

  When a user access the agent, they'll be prompted for (Entra) authorization, and then Gemini Enterprise will pass the user's access token to the agent. The agent can use that token to identify the user (by inspecting the token's claims) or make calls to Microsft APIs.

• To have Gemini Enterprise pass a workforce identity federation token, you can use a helper tool, AAAuth. The workforce identity federation token won't tell the agent who the user is, but the agent can use that token to access Google APIs as the end user.