adk CLI is how you scaffold, run, deploy, and debug ADK agents. Run adk --help for the top-level command list, or adk <command> --help for flags on a specific command.
Global flags
These flags work with every command:
Most commands also accept
--format json for machine-readable output. The following commands don’t: dev, login, profiles list, profiles set, upgrade, remove, self-upgrade, telemetry, theme, mcp, mcp:init, run, assets pull.
Project
These commands manage an agent project from scaffold to deploy:adk init
Scaffold a new agent project. Pass a name, or omit it to be prompted:
adk dev
Start the dev server with hot reload:
adk deploy
Deploy the built agent to Botpress Cloud:
adk link
Link the local agent to a workspace and bot. Writes to agent.json, or to agent.local.json with --local:
Integrations
These commands add, inspect, and manage the integrations your agent depends on:adk add
Add an integration or interface to your agent. Accepts a name, workspace/name, or a specific version:
adk search
Search the Botpress Hub for integrations:
adk list
List integrations in your project, or all available ones on the Hub:
adk info
Show details for a specific integration. Filter by a single facet or show the full spec:
Configuration and secrets
These commands manage config values and secrets:
All configuration and secret commands accept
--prod to target the production environment instead of dev.
Chat and testing
These commands let you send messages to your agent and run eval suites:adk chat
Open an interactive chat, or send a single message with --single. Requires adk dev to be running:
adk evals
Run eval suites. With no arguments, runs all evals; pass a name to run just one:
adk evals runs
List past eval runs, or show the details of a specific one:
Workflows
These commands discover and run workflows against the local dev server:adk workflows run
Run a workflow with a JSON payload. Add --wait to block until it finishes:
Debugging
These commands help you inspect what your agent is doing:adk logs
Query logs from .adk/logs/. Pass filter tokens as positional arguments, or stream with --follow:
adk traces
Query trace data from the local store. Filter tokens narrow by workflow, action, trigger, or drill into a specific trace:
adk run
Run a TypeScript script with the full ADK runtime, so you can import actions, Zai, and tables directly:
Knowledge bases
Sync local knowledge base content with Botpress:adk kb sync
Push local KB sources to the dev or prod bot. Use --dry-run to preview changes first:
Assets
Manage static files your agent serves or references:adk assets sync
Upload local assets to remote storage. Use --dry-run to see what would change:
adk assets list
List asset files. By default shows both local and remote; filter with --local or --remote:
AI assistants
These commands integrate the ADK with AI coding tools:adk mcp
Start the MCP server so tools like Claude Code, Cursor, or VS Code can talk to your running dev server:
adk mcp:init
Generate MCP configuration files for supported AI assistants:
Account
These commands manage your Botpress credentials and the CLI itself:adk login
Authenticate with your Botpress account. Pass --token to skip the browser flow:
adk self-upgrade
Upgrade the ADK CLI. Pass a tag (beta, next) or an explicit version: