Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    ADK Multi-Agent Systems on Cloud Run

    1

    Ship Google ADK agents and multi-agent systems to Cloud Run without the classic failures: missing root_agent, 403s to private MCP servers, lost sessions, CORS *, and GPUs you never needed.

    $9

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    ADK Multi-Agent Systems on Cloud Run

    ADK Multi-Agent Systems on Cloud Run

    Example session with this skill installed

    We're about to deploy our ADK sales agent (in service/) to Cloud Run with deploy.sh. It calls our maps MCP server, another Cloud Run service. Last week's test deploy cost a fortune and the agent kept forgetting conversations. Review everything and write the fixes in priority order in REVIEW.md.

    • Read your context and instructions
    • Compiled the adk multi-agent systems

    REVIEW.md: 1) Hard-coded Maps key in agent.py and deploy.sh: move to Secret Manager and rotate. 2) MCP service is public: deploy with --no-allow-unauthenticated, grant the agent's service account run.invoker, send an ID token with the service URL as audience. 3) Forgotten conversations: SQLite sessions are per instance; use a database or managed session service. 4) Cost: drop the GPU and 8 CPU / 32 GiB; Gemini runs at Google, start with 2 CPU / 2 GiB. 5) agent.py defines sales, not root_agent, and has no init.py, so ADK loads nothing. 6) CORS "*" and web=True. 7) Pin google-adk; raise timeout from 60 s to 300 s; min-instances 1.

    REVIEW.md

    Connects securely to your tools. The creator never sees your data.

    About this skill

    ADK makes the first agent easy. Production breaks on the parts around it: the server cannot find the agent package, the MCP server answers 403 because the call carries no identity, sessions vanish on every new instance, CORS is *, the service runs on GPUs it never uses, and the "multi-agent" design is four agents doing one agent's job.

    What your agent does with it

    • Chooses the least machinery: single agent + tools, specialists as tools (AgentTool), sub-agents, workflow agents, or separate A2A services, with clear rules for each.
    • Lays out the project so get_fast_api_app and adk web actually load it (root_agent, __init__.py, Dockerfile on $PORT).
    • Wires private MCP servers as their own Cloud Run services, with run.invoker for the agent's service account and ID tokens with the right audience.
    • Configures Vertex AI and durable sessions (SQLite in a container forgets every conversation on scale-out).
    • Right-sizes Cloud Run: no GPUs for agents that call Gemini over the API, timeouts for multi-step turns, warm instances, secrets from Secret Manager, explicit CORS.
    • Runs a static preflight (adk_preflight.py) that fails CI on a missing root_agent, allow_origins=["*"], SQLite sessions, key-shaped strings in code, unpinned google-adk.
    • Generates a reviewed deploy plan (cloudrun_plan.py): service accounts, roles, invoker binding, flags and a smoke test, with warnings for risky choices.

    Included: SKILL.md, code patterns (serving, specialists, MCP with ID tokens, A2A, Live agents, smoke test), both scripts (standard library, they never run gcloud), and an eval case.

    Part of the Production AI Agents series. Read Agent Architecture (free) first; pair with MCP Server in Production.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.

    Cloud Run only (not GKE or Agent Engine deployment in depth). Python ADK only. Code patterns are dated to ADK 1.x in 2026 and should be checked against the version you pin.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 day ago

    Software
    Python 3.10+Google Cloud project with billing
    Optional
    gcloud CLIDocker

    Permissions

    File Scopes

    Your
    service
    folder
    (read
    by
    the
    preflight)
    review
    and
    plan
    files
    you
    ask
    for

    adk_preflight.py only reads files; cloudrun_plan.py prints gcloud commands and never runs them.

    Listed1 day ago

    What's inside

    Frequently Asked Questions

    No. cloudrun_plan.py prints a reviewed sequence of commands with warnings; you run them.

    Usually not. The skill starts from one agent with good tools and adds specialists only for context isolation or different permissions.

    An agent calling Gemini over the API does no model inference in your container. GPUs only make sense for self-hosted models and multiply the cost.

    New in LLM Ops