A powerful Python API, CLI, and set of Agent Skills for CX Agent Studio to automate, evaluate, and scale your agents with ease.
Important Links: Docs, Examples, Agent Skills, Core SDK
CX Agent Studio Scripting API (CXAS SCRAPI) is an open-source Python scripting API, CLI, and set of Agent Skills for CX Agent Studio. It is designed to simplify building, deploying, and orchestrating agent workflows, from simple tasks to complex systems. It integrates seamlessly with Agentic IDEs like Gemini CLI, Claude Code, and Antigravity, exposing advanced tooling for deep evaluations, real-time latency metrics, offline linting, and conversation history.
- Python 3.10+
Authentication can vary depending on how and where you are interacting with SCRAPI.
If you're using CXAS SCRAPI with a Google Colab notebook, you can add the following to the top of your notebook for easy authentication:
pip install cxas-scrapiproject_id = '<YOUR_GCP_PROJECT_ID>'
# this will launch an interactive prompt that allows you to auth with GCP in a browser
!gcloud auth application-default login --no-launch-browser
# this will set your active project to the `project_id` above
!gcloud auth application-default set-quota-project $project_idAfter running the above, Colab will pick up your credentials from the environment and pass them to CXAS SCRAPI directly. No need to use Service Account keys! You can then use CXAS SCRAPI simply like this:
from cxas_scrapi import Apps
project_id = '<YOUR_GCP_PROJECT_ID>'
location = 'us'
app_client = Apps(project_id=project_id, location=location) # <-- Creds will be automatically picked up from the environment
apps_map = app_client.get_apps_map()If you're using CXAS SCRAPI with Cloud Functions or Cloud Run, CXAS SCRAPI can pick up on the default environment creds used by these services without any additional configuration!
- Add
cxas-scrapito yourrequirements.txtfile - Ensure the Cloud Function / Cloud Run service account has the appropriate Custom Agent / Conversational Agents IAM Role
Once you are setup with the above, your function code can be used easily like this:
from cxas_scrapi import Agents
app_name = '<YOUR_APP_NAME>'
a = Agents(project_id='<YOUR_GCP_PROJECT_ID>', location='global')
agents_map = a.get_agents_map(app_name)Similar to Cloud Functions / Cloud Run, CXAS SCRAPI can pick up on your local authentication creds if you are using the gcloud CLI.
- Install gcloud CLI.
- Run
gcloud init. - Run
gcloud auth login - Run
gcloud auth application-default login - Run
gcloud auth listto ensure your principal account is active.
This will authenticate your principal GCP account with the gcloud CLI, and SCRAPI can pick up the creds from here.
If you prefer to explicitly assign Service Account credentials programmatically instead of relying on the environmental application-default, you can pass the path to your JSON key using creds_path.
from cxas_scrapi import Tools
creds_path = '<PATH_TO_YOUR_SERVICE_ACCOUNT_JSON_FILE>'
t = Tools(project_id='<YOUR_GCP_PROJECT_ID>', location='global', creds_path=creds_path)
tools_map = t.get_tools_map('<YOUR_APP_NAME>')Set up Google Cloud Platform credentials and install dependencies.
gcloud auth login
gcloud auth application-default login
gcloud config set project <project name>uv sync
source .venv/bin/activateTo run a simple bit of code you can do the following:
- Import a Class from
cxas_scrapi - Define your GCP Project and Location
from cxas_scrapi import Apps
# Instantiate your class object and pass in your credentials
app_client = Apps(project_id='<YOUR_GCP_PROJECT_ID>', location='global')
# Retrieve all Apps existing in your project
apps = app_client.list_apps()
for app in apps:
print(app.display_name, app.name)Here is a brief overview of the CXAS SCRAPI library's structure and the motivation behind that structure.
The src/cxas_scrapi/core directory contains the high level building blocks of CXAS SCRAPI, mapped to core resource types in the CXAS environment (Apps, Agents, Tools, Guardrails, Deployments, Sessions, etc.)
The src/cxas_scrapi/utils directory contains helper functions and background logic for pagination, response flattening, proto conversions, and external integrations like Google Sheets.
The src/cxas_scrapi/evals directory provides tools for executing and analyzing agent performance evaluations, including Golden tests and simulation runs, and extracting metrics like latency.
The src/cxas_scrapi/cli directory implements the command line interface for SCRAPI, offering tools like cxas lint to automate development and validation workflows.
The src/cxas_scrapi/migration directory contains tools to facilitate transitioning legacy Dialogflow CX agents to CXAS, including agent generation from flows and artifact building.
The official documentation is hosted online at https://googlecloudplatform.github.io/cxas-scrapi/stable/.
The documentation site is built with MkDocs Material. To run it locally:
# Install dependencies including docs extras
uv sync --extra docs
# Start the local dev server
uv run mkdocs serveOpen http://127.0.0.1:8000 in your browser. Changes to files in docs/ will reload automatically.
To build the static site without serving:
mkdocs build # output goes to site/
mkdocs build --strict # also fails on warnings (used in CI)We welcome any contributions or feature requests you would like to submit!
- Fork the Project
- Create your Feature Branch (git checkout -b feature/AmazingFeature)
- Commit your Changes (git commit -m 'Add some AmazingFeature')
- Push to the Branch (git push origin feature/AmazingFeature)
- Open a Pull Request
Distributed under the Apache 2.0 License. See LICENSE for more information.
Patrick Marlow - pmarlow@google.com - @kmaphoenix
Project Link: https://github.com/GoogleCloudPlatform/cxas-scrapi