Claude Code Setup for PPC Agency Teams: Shared Workspace Guide

Gartner found 65% of marketing agencies now deploy at least one AI tool. Most still lack a setup that survives five AMs and twelve clients. That is the gap this guide closes.

TL;DR

• Most Claude Code setup guides assume one practitioner, one laptop, one client. Agency teams need a shared workspace, multi-account API strategy, and team-wide prompt governance, which none of the solo guides cover.
• The critical pre-step is deciding your MCC-to-developer-token strategy: a single token under an MCC manager account can reach all sub-accounts you manage, avoiding per-client OAuth setup.
• An agency CLAUDE.md file is the difference between Claude Code producing generic output and output that matches your agency's naming, quality, and client-review standards every time.
• Shared prompt libraries version-controlled in Git give every team member the same automation toolkit, no more "can you send me that script you ran last week."
• The most common agency mistake is skipping the sandbox validation step, which causes junior team members to accidentally push unvalidated changes to live client accounts.

If you run PPC at an agency, you have already read one of the solo-practitioner Claude Code guides. You installed the terminal. You got the developer token. You pasted the kickoff prompt. And then you hit the wall that every agency team hits: How do I roll this out across five account managers, twelve clients, and a shared codebase without chaos? Most shops still lack a repeatable claude code setup for ppc agency teams.

The solo guides work fine when one person automates their own workflows. But claude code setup for ppc agency teams is a different problem. You need multi-account API access that doesn't break when a team member leaves. You need a CLAUDE.md that enforces your agency's naming conventions, ad copy quality rules, and review-before-publish gates. You need a shared prompt library so your junior analyst runs the same keyword expansion script your senior strategist built last month, not a half-remembered version.

This guide covers exactly that: the step-by-step claude code setup for ppc agency teams, built for multiple people, multiple clients, and real agency operations. A claude code setup for ppc agency teams is an infrastructure project. Not a weekend toy.

What Changes When a PPC Agency Team, Not Just One Practitioner, Adopts Claude Code

The solo practitioner workflow is simple: one machine, one Google Ads account (or a few linked under an MCC), one person who knows where everything lives. When something breaks, the same person who wrote the script fixes it. When the prompt needs updating, the same person who created it edits it. A proper claude code setup for ppc agency teams must account for three new dimensions that the solo guides ignore entirely.

An agency team introduces three new dimensions:

Account separation. Your team manages 10, 20, or 50 client accounts. If your Claude Code setup ties every script to a single developer token and a single OAuth credential, you cannot safely switch between clients without manually swapping config files, a process that will be forgotten on a busy Monday, and a change meant for Client A will land in Client B's account.

Team onboarding and offboarding. A senior strategist builds a powerful campaign-analysis prompt. Two weeks later they leave the agency. Where is that prompt? In their local terminal history? In a Slack DM they sent to the founder? An agency-grade setup treats prompts and scripts as team assets, not personal files.

Quality and safety gates. Solo practitioners know exactly what they asked Claude Code to do. When a junior account manager runs the same automation, they may not recognize when the output looks wrong, or worse, when Claude Code proposes a change that violates a client's targeting restrictions. Your setup needs guardrails that don't require senior judgment for every single action.

• Agency setups must separate client credentials at the environment level, not the script level.
• Team members should inherit automation tooling from a shared source (GitHub repo, team drive) rather than duplicating and drifting.
• Validation steps (dry-run modes, approval gates, change logging) are non-negotiable when multiple people share access to live ad accounts.

A PPC agency that treats Claude Code as "one person's automation toy" captures maybe 20% of the tool's value. A team that treats it as a shared automation layer multiplies that across every account and every team member. That is the core argument for treating your claude code setup for ppc agency teams as an infrastructure project, not a side experiment.

The Agency Prerequisites You Can't Skip Before Your Claude Code Setup for PPC Agency Teams

Before you install anything, make these three decisions as a team. Changing them later means re-doing your entire claude code setup for ppc agency teams.

1. Multi-Account API Strategy

Every Google Ads API developer token is originally issued to a single account. If you manage client accounts under an MCC (Manager Account), you can use a single developer token from that MCC to access every linked sub-account. This is the recommended approach for agencies. It is the MCC as API identity model. Pair it with the same margin logic as AI marketing agents for PPC agencies.

If instead each team member uses their own Google Ads account's token, you multiply OAuth setup by the number of people and the number of accounts. That becomes unmanageable fast.

> Decision: Appoint one MCC account as the "API identity" for your team. Generate the developer token and OAuth credentials there. Every Claude Code session on every team machine uses the same MCC-level token. Account switching happens via the login_customer_id parameter in the API call, not by swapping credentials.

2. Shared Workspace and Version Control

Decide where your team's Claude Code workspace lives. Options:

The last row is the tell. Per-person copies are how Client A changes land in Client B.

Recommendation: A private GitHub repo with a branching convention. Main branch holds the team's approved CLAUDE.md and shared prompt library. Individuals branch for client-specific experiments.

3. Plan Tier Decision

Claude Code requires a paid Anthropic plan. The Pro plan includes Claude Code access; the Team plan adds centralized billing and admin controls. For agencies with 3+ users, the Team plan saves administrative overhead.

Cost is not the decision. Admin overhead is. Three or more users should not split five credit cards.

Step-by-Step: Claude Code Setup for PPC Agency Teams

The steps below assume you have chosen your MCC strategy and shared workspace approach from the prerequisites. Each step is designed specifically for a claude code setup for ppc agency teams, they differ from the solo guides in every meaningful way.

Step 1, Standardize the Workspace Structure

Create the shared workspace directory and initialize it as a Git repo on one machine, then push it for the team to clone.

mkdir agency-ppc-automation
cd agency-ppc-automation
git init

Inside this folder, create the skeleton:

agency-ppc-automation/
├── CLAUDE.md              # Team-wide instructions for Claude Code
├── prompts/               # Shared prompt library
│   ├── account-audit.md
│   ├── keyword-expansion.md
│   ├── negative-keyword-extract.md
│   └── reporting-template.md
├── scripts/                # Python/JS scripts Claude Code may call
├── clients/                # Per-client config files (gitignored)
│   ├── client-a.env
│   └── client-b.env
└── sandbox/                # Test account workspace

This skeleton belongs in the shared GitHub repo. Every team member clones it once.

Step 2, Set Up Google Ads API Access for Multiple Accounts

From your MCC account, generate the developer token and set up OAuth. Use the Google Ads API docs to create a project, enable the Google Ads API, and create OAuth 2.0 Desktop credentials.

Claude Code can walk you through this interactively. Paste this kickoff prompt into Claude Code:

I need to set up Google Ads API access for an agency managing multiple client accounts from one MCC. I already have my developer token from the API Center. Walk me through creating OAuth credentials in Google Cloud Console, generating a refresh token, and testing access to a sub-account. I'm not a developer.

The tool will create a Python script that runs the OAuth flow and stores the refresh token. Store this token in a shared but gitignored config file (.env or similar) so each team member can use it without exposing credentials in version control.

For each client account, add a client-X.env file:

GOOGLE_ADS_DEVELOPER_TOKEN=your_token
GOOGLE_ADS_CLIENT_ID=your_client_id
GOOGLE_ADS_CLIENT_SECRET=your_client_secret
GOOGLE_ADS_REFRESH_TOKEN=your_refresh_token
GOOGLE_ADS_LOGIN_CUSTOMER_ID=123-456-7890

Step 3, Write the Agency CLAUDE.md

This file is your team's instruction manual for Claude Code. Every session starts with it. Make it thorough. A well-written CLAUDE.md is the single highest-ROI step in any claude code setup for ppc agency teams because it encodes your agency's standards once and applies them to every session.

Include:

• Naming conventions: "Use sentence case for headlines. Label all test campaigns with [TEST] prefix."
• Quality rules: "Never change a campaign budget without logging the reasoning. Never pause a campaign with active conversions without asking."
• Client separation: "When working on a client account, always read the corresponding clients/client-name.env file first. The login_customer_id in that file determines which account you are affecting."
• Approval gates: "Before making changes to any live account, run /google-ads:validate-changes and send the output to the senior account manager."
• Prompt sourcing: "Check prompts/ before writing a new prompt from scratch. If you create a new prompt that others can use, add it to the library."

A good CLAUDE.md transforms Claude Code from a generic coding tool into an agency-specific automation assistant that follows your team's standards automatically.

Step 4, Build Your Shared Prompt Library

Prompts are the reusable automation templates your team runs day to day. Store them as plain markdown files in prompts/. Here is an example prompt file for a weekly campaign audit:

# Campaign Health Audit
**Purpose:** Quick health check on all active campaigns in a client account.
**Steps:**
1. Pull campaign performance for last 14 days.
2. Flag any campaign with CTR < 1% or conversion rate drop > 20% WoW.
3. Suggest 2, 3 optimizations per flagged campaign.
4. Output results as a formatted report suitable for client email.

Those four files are the minimum library. They are enough for week one. A fifth file can wait until the first four run weekly.

Team members run a prompt by pasting its content into Claude Code (or referencing it via a prompt-loading script). The key is that everyone runs the same prompt, not "the version Sarah tweaked last Thursday."

Step 5, Validate Against a Sandbox Account During Your Claude Code Setup for PPC Agency Teams

Before any team member runs a prompt on a live client account, they should test it against a Google Ads sandbox or a small, low-spend test account. This validation step is what separates a safe claude code setup for ppc agency teams from a risky one.

The sandbox pattern:

1. Create a free Google Ads test account (Google provides these for API development).
2. Set GOOGLE_ADS_LOGIN_CUSTOMER_ID to the test account ID.
3. Run the prompt against that test ID, then have a senior read the output for unexpected change proposals.
4. Only then switch to the live client's customer ID.

This is the step most agency teams skip, and it is the step that prevents "I accidentally paused the wrong campaign" scenarios.

How Agency Teams Use Claude Code Day-to-Day

Once your claude code setup for ppc agency teams is running, three patterns emerge that solo practitioners never encounter.

The Reporting Template Pattern

An account manager opens Claude Code on Monday morning, selects the client's .env file, and runs the reporting-template prompt. Claude Code pulls the last 7 days of data, formats it into the agency's standard client report structure, and outputs a markdown draft. The AM reviews, adjusts the narrative, and sends. Every week, every client, the same structure, no more rebuilding pivot tables from scratch.

The Shared Prompt Library Pattern

A senior strategist discovers a new optimization, say, identifying campaigns where Impression Share lost to budget exceeds 15%. They write the prompt once, test it, and add it to prompts/. The next day, every account manager on the team can run the same analysis on every client. The agency's institutional knowledge compounds instead of walking out the door.

The Junior QA Pattern

A junior team member runs the negative-keyword-extract prompt on a new client account. Claude Code returns a list of candidate negatives. The junior reviews the list in the tool, flags anything ambiguous, and sends the final list to the senior for approval, all without touching the live account. The approval gate is built into the workflow, not enforced by trust. This is the same architecture AI marketing agents for PPC agencies use for observational vs execution agents. SEO shops hit the same credential split in AI marketing agents for SEO agencies. Content teams do it in AI marketing agents for content agencies.

Five Mistakes PPC Agency Teams Make During Their First Claude Code Setup

These mistakes share one root. The team treated Claude Code as a personal toy. It is infrastructure. Read the table as a pre-mortem before you give a junior a live token.

The first row is the one that produces Client A changes in Client B. Fix that before you write a clever prompt. The sandbox row is how juniors stay employed. If you cannot name who owns the MCC token, you are not ready for live execution. That owner belongs in CLAUDE.md, and the repo gets pulled every Monday.

PPC retainers leak into unbillable search-term hours. Clients still buy judgment. They do not buy a junior with a live token and no sandbox. You already know which three accounts would notice a silent pause. You already know which AM still keeps prompts in Notes.app.

The missing layer is a workflow that drafts the change, waits for a human, then compounds that judgment into a reusable skill. Skills, context, and agents only pay off when the next retainer inherits the last retainer's lessons. Metaflow is built for that compounding loop: the same operator can encode MCC identity, CLAUDE.md gates, and sandbox checks once, then run them as workflows across the book without cloning a brittle sheet for every client.

Frequently Asked Questions

Can Claude Code access multiple Google Ads accounts from one setup?

Yes. When you configure Claude Code with a developer token issued by an MCC account, you can switch between sub-accounts by changing the login_customer_id parameter. Each client gets its own .env file with the appropriate ID. Claude Code reads that file to determine which account to query or modify. Metaflow teams encode that MCC as API identity in the workflow, not as a sticky note on one laptop.

Do I need an MCC to use Claude Code for multiple clients?

You do not strictly need an MCC, each Google Ads account can generate its own developer token. But for agency teams managing more than three client accounts, an MCC is the practical choice. It gives you one developer token, one OAuth credential set, and centralized access management across all sub-accounts. Metaflow teams treat that MCC as API identity as a named choice, not as whatever laptop was nearby.

Is Claude Code safe for client account data?

Claude Code runs locally on your machine. No client data is sent to Anthropic's servers beyond what you explicitly ask Claude Code to process through the API. For additional safety, use the dry-run commands available in the Google Ads plugin for Claude Code (such as /google-ads:validate-changes) to review every change before it executes. No change hits a live account without your explicit approval. A claude code setup for ppc agency teams that skips dry-run is not a setup. It is a live wire.

How much does Claude Code cost for an agency team?

Claude Code is included in Claude Pro ($20/month per person) and Claude Team ($25/month per person). For a team of five, that is $100, 125/month, roughly the cost of one hour of a senior PPC manager's time. The API-based billing model adds additional charges for high-volume usage but is not required for most agency workflows. Google's own bidding overview is still the source of truth for what the API will let you change. Metaflow's bet is that the workflow and the context layer matter more than the logo on the login screen.

Can junior team members use Claude Code safely?

Yes, with the right guardrails. Require sandbox testing before live execution. Enforce the /google-ads:validate-changes step through your CLAUDE.md. Use the plan-then-execute workflow (Plan Mode) so junior team members review proposed changes before applying them. The tool itself is forgiving, errors can be pasted back into Claude Code for it to fix.