ADK (Agent Development Kit) to connect to over 100+ Applications

Google Cloud Integration Services
,
1

In case you haven’t heard the latest announcement about Google Cloud’s Agent Development Kit (ADK), you have to check out one of the coolest features embedded in the ADK. The ability to connect through Application Integration’s Integration Connectors to over 100+ different kinds of applications.

Please see this awesome medium blog, written by @lolejniczak17 that takes you through, step by step, how to do this.

https://medium.com/google-cloud/connect-act-google-adk-agents-with-gcp-integration-connectors-to-perform-tasks-across-100-ca10a3fd5334

5 Likes
2

I used the article to build aAgent using the Agent Development Kit (ADK), thn tried to deployed it it’s a Reasoning Engine (agent engine).

The agent uses a tool that invokes an Application Integration connector (ListFiles) to access Google Drive. The agent deploys successfully, and I’m able to create a session without issues.

However, when I try to send a message to the agent, I get the following error:

PERMISSION_DENIED: Permission 'integrations.integrations.invoke' denied on resource '//integrations.googleapis.com/projects/project_id/locations/europe-central2/integrations/ListFiles'

I created a service account for my integration application and gave it this role:

roles/integrations.integrationInvoker

I also gave the following role to the agent engine service account:

roles/iam.serviceAccountTokenCreator to service-[project-id]@gcp-sa-aiplatform.iam.gserviceaccount.com

But I am still getting this error and I don’t understand the problem any idea?

4

Dependency Conflict: ResolutionTooDeep Error with ADK, google-cloud-aiplatform, and other Google AI Libraries

Hello, ADK Community,

I am encountering a build failure in my CI/CD pipeline and am hoping someone here might have experience with a similar issue. The build process, which uses a standard pip install -r requirements.txt command inside a Docker container, consistently fails with a pip._vendor.resolvelib.resolvers.ResolutionTooDeep: 200000 error.

This indicates that pip’s dependency resolver is unable to find a compatible set of packages after a massive number of backtracking attempts. My analysis suggests there’s an unresolvable conflict between several high-level packages in my requirements.txt file, particularly concerning shared foundational libraries.

Here is my complete requirements.txt:

google-cloud-secret-manager
google-cloud-aiplatform==1.101.0
cloudpickle>=2.0.0
vertexai
anthropic
configobj
configparser
gcsfs==2021.11.0
GitPython==3.1.43
google_adk==1.4.2
google-cloud-logging
grpcio-status
litellm
nipype
pip
PySocks
pyxnat
selenium>=4.0.0
pathlib
pyyaml
jira==3.8.0
pandas
fastapi
openai==1.95.1
google_generativeai==0.7.2
google-cloud-storage>=2.10.0
tenacity>=8.2.3
google-api-core>=2.11.0
langchain==0.3.27
langchain-core==0.3.74
langwatch==0.2.17

The conflict appears to be rooted in the incompatible version constraints that packages like google-cloud-aiplatform, google_adk, and langwatch place on common dependencies, such as protobuf, grpcio, and google-api-core.

I have tried running pip install in a clean environment, but the error persists. The extensive backtracking suggests that there is no single version of these core libraries that satisfies the requirements of all the packages simultaneously.

My specific question to the community is:

  • Has anyone successfully used a combination of google-adk==1.4.2, google-cloud-aiplatform==1.101.0, langchain==0.3.27, and langwatch==0.2.17 in a single project? If so, could you share the compatible versions of the key transitive dependencies like protobuf, grpcio, or google-api-core that you used to resolve this?

  • Are there any known best practices or a recommended set of versions for these packages to work together harmoniously?

Any guidance or a known working configuration would be greatly appreciated. Thank you in advance for your help!

5

Thanks for sharing! The ADK’s ability to connect with 100+ applications through Integration Connectors is a game-changer for streamlining workflows. For anyone looking to automate tasks across multiple apps in Google Cloud, that Medium guide is definitely worth a read—makes setup way easier.

1 Like
6

Thanks for engaging in our forums! For more tailored guidance, I recommend sharing your question in this other forum: Agents - Google Developer forums

I hope this helps!

7

Hey Everyone,

Does anyone have experience handing over control from a Google ADK agent to a remote agent using the A2A protocol? I’m trying to move the conversation from the bot to a secondary tool/human agent, but I’m not sure how to signal the ‘handover’ event correctly. Any tips on getting the remote agent to take full control of the session? Please do let me know if you find the answer for it.