Claude Code Agent Teams for Marketers: Full-Cycle GTM Playbook with Parallel AI Agents

Claude Code agent teams let you run multiple AI agents in parallel on the same marketing project. One agent analyzes your product usage data for churn signals. Another builds your go-to-market plan. A third writes landing page copy while a fourth generates ad creatives through Canva. A fifth sets up GA4 event tracking. They coordinate, share findings, and report back to a team leader. All at the same time.

This is not a “how to set up agent teams” guide. This is a full-cycle GTM playbook. You will walk through every stage of marketing operations, from product intelligence through campaign reporting, using parallel AI agents connected to your martech stack via MCP servers. Every section includes copy-paste prompts, MCP configurations, and time-saved estimates based on a concrete example: launching a retention campaign for a fictional B2B SaaS analytics tool called “MetricFlow.”

TL;DR

Claude Code agent teams run parallel AI agents across your full go-to-market cycle. This playbook covers seven stages: product data analysis, GTM strategy, channel selection, content production (blog, landing page, email, social, newsletter), ad creative generation via Canva MCP, GA4 event tracking setup, and automated weekly reporting. Each stage includes agent team prompts, MCP server configs, and time-saved estimates.

Total savings: 25-35 hours per campaign cycle using parallel agents connected to Semrush, HubSpot, Google Analytics 4, Canva, and Figma.

Key Takeaways

• Agent teams run parallel Claude Code sessions with independent context windows
• Full GTM cycle compresses from 2-3 weeks to 2-3 days
• MCP servers connect agents to Semrush, HubSpot, GA4, Canva, Figma, and Slack
• Opus for the team leader, Sonnet for all teammates controls costs
• Skills convert agent team workflows into reusable one-line commands
• Every stage includes a copy-paste prompt tested on the MetricFlow example
• Full campaign setup costs $15-25 in tokens

What Are Claude Code Agent Teams?

Agent teams let you coordinate multiple Claude Code sessions working together on a shared project. One session acts as the team leader. It plans the work, creates tasks, spawns teammates, and synthesizes results. Each teammate operates in its own context window with its own conversation history.

The difference from sub-agents is communication. Sub-agents work within a single session and report results only to the parent agent. Agent team members message each other directly, share findings mid-task, and coordinate without the leader acting as intermediary. When the research agent finds a competitor positioning gap, it messages the copywriter agent directly.

Agent teams shipped with the Opus 4.6 release in February 2026 as a research preview. The feature is experimental but functional. Anthropic’s own growth marketing team already uses specialized sub-agents to generate hundreds of ad variations in minutes. Agent teams extend this pattern to full campaign orchestration.

Action item: Open the agent teams documentation and review the use case examples before continuing.

How Do You Set Up Agent Teams and MCP Servers?

Step 1: Update Claude Code and Enable Agent Teams

Agent teams require the latest Claude Code version and two configuration flags. Run these commands in your terminal:

claude update
claude config set -g agentTeams true
claude config set -g tmuxSplitPanes true

Step 2: Install tmux for Split Pane View

Tmux displays each teammate in a separate terminal pane. Without it, all agents output into one noisy stream.

# macOS
brew install tmux

# Ubuntu/Debian
sudo apt install tmux

Tmux does not work inside VS Code or Cursor terminals. Use your native system terminal.

Step 3: Connect Your Marketing MCP Servers

MCP servers give your agents live access to marketing platforms. Each agent on the team inherits MCP server connections from your Claude Code configuration. Connect these servers before spawning any agent team.

Semrush MCP (SEO data, keyword research, competitor analysis):

claude mcp add semrush https://mcp.semrush.com/v1/mcp -t http

HubSpot MCP (CRM data, contact records, deal pipeline):

claude mcp add --transport http hubspot --scope user https://mcp.hubspot.com/anthropic

GA4 MCP by Google (official analytics data, traffic sources, events):

# Install via pipx (requires Python 3.10+)
pipx install analytics-mcp

# Authenticate with Google Cloud ADC
gcloud auth application-default login \
  --scopes https://www.googleapis.com/auth/analytics.readonly,https://www.googleapis.com/auth/cloud-platform \
  --client-id-file=YOUR_CLIENT_JSON_FILE

# Add to Claude Code
claude mcp add --transport stdio analytics-mcp -- pipx run analytics-mcp

Google’s official GA4 MCP server (github.com/googleanalytics/google-analytics-mcp) connects to both the GA4 Reporting API and Admin API. Read-only access. Requires a Google Cloud project with the Analytics Data API and Admin API enabled, plus Application Default Credentials (ADC) configured.

Figma MCP (design files, component extraction, layout data):

claude mcp add --transport http figma https://mcp.figma.com/mcp

Canva MCP (design creation, ad creatives, on-brand assets):

claude mcp add canva-dev -- npx -y @canva/cli@latest mcp

Slack MCP (team notifications, reporting):

claude mcp add --transport stdio slack -- npx -y @modelcontextprotocol/server-slack

Verify all connections with the /mcp command inside Claude Code. Each server should show as “connected” with its tools listed.

Step 4: Start Your tmux Session

Always start tmux before launching Claude Code:

tmux new -s campaign
claude --dangerously-skip-permissions

Select Opus 4.6 as your model. Set effort to low for initial planning, then increase for complex tasks.

Action item: Run through all four setup steps. Connect at minimum Semrush and Google’s official GA4 MCP servers. Run /mcp to confirm both show as connected.

The MetricFlow Campaign: Full-Cycle GTM Example

Every prompt in this playbook targets one scenario: MetricFlow, a fictional B2B SaaS analytics tool, needs to launch a retention campaign. Usage data shows 23% monthly churn among mid-market accounts. The marketing team has one operator (you) with Claude Code agent teams.

Here is the seven-stage workflow:

1. Product data analysis (identify churn patterns and retention levers)
2. GTM strategy (build the retention campaign plan)
3. Channel selection (map channels to audience segments)
4. Content production (blog post, landing page, emails, social posts, newsletter)
5. Ad creative generation (display ads and social ads via Canva MCP)
6. Event tracking setup (GA4 events, UTM schema, conversion tracking)
7. Automated reporting (weekly performance dashboards and Slack alerts)

Each stage uses a dedicated agent team. Some stages use two to three agents. Others use five to six. The total cost across all seven stages: $15-25 in API tokens.

Stage 1: How Do You Analyze Product Data for Campaign Intelligence?

Product usage data tells you where churn happens. Agent teams analyze this data from multiple angles simultaneously, eliminating the anchoring bias of a single-agent investigation.

The Agent Team Prompt

SYSTEM: You are a product analytics team leader for MetricFlow, a B2B SaaS analytics platform.

Create an agent team to analyze our product usage data and identify churn patterns for a retention campaign.

<context>
Product: MetricFlow - B2B SaaS analytics tool for mid-market companies
Problem: 23% monthly churn among mid-market accounts (50-500 employees)
Data location: Export files in data/usage-export.csv and data/churn-cohorts.csv
CRM: HubSpot (connected via MCP)
</context>

Spawn these teammates:

- Usage Pattern Analyst: Analyze usage-export.csv. Identify which features correlate with retention vs. churn. Find the "aha moment" threshold (feature adoption level where churn drops below 10%). Save findings to campaigns/retention/analysis/usage-patterns.md.

- Churn Investigator: Analyze churn-cohorts.csv. Segment churned accounts by company size, industry, onboarding completion rate, and time-to-first-value. Identify the top 3 churn triggers. Save to campaigns/retention/analysis/churn-triggers.md.

- CRM Intelligence Agent: Query HubSpot via MCP. Pull deal stage data, support ticket history, and NPS scores for churned vs. retained accounts. Identify leading indicators of churn visible in CRM. Save to campaigns/retention/analysis/crm-signals.md.

- Retention Strategist: Wait for all other agents to complete. Synthesize all findings into a retention campaign brief with specific messaging angles, target segments, and recommended interventions. Save to campaigns/retention/analysis/campaign-brief.md.

Use Sonnet for all teammates.

Churn Investigator and Usage Pattern Analyst MUST message each other when they find overlapping patterns.
Retention Strategist reads all three analysis files before writing the brief.

What This Team Produces

Four agents work in parallel for 10-15 minutes. The Usage Pattern Analyst identifies feature adoption thresholds. The Churn Investigator finds segment-specific triggers. The CRM Intelligence Agent pulls live HubSpot data. The Retention Strategist waits for all inputs, then synthesizes a campaign brief with specific messaging angles used in every downstream stage.

Time saved: 8-12 hours of manual data analysis compressed to 15 minutes. Cost: $2-3 in tokens.

Action item: Export your product usage data and churn cohort data to CSV files. Place them in a data/ directory in your Claude Code project. Create the campaigns/retention/analysis/ directory structure.

Stage 2: How Do You Build a Go-To-Market Plan with Agent Teams?

The campaign brief from Stage 1 feeds directly into GTM planning. Agent teams build the plan from competing perspectives, preventing the single-viewpoint bias of sequential planning.

The Agent Team Prompt

SYSTEM: You are a GTM strategy team leader for MetricFlow.

Create an agent team to build a retention campaign go-to-market plan.

<context>
Read campaigns/retention/analysis/campaign-brief.md for the full campaign brief.
Budget: $15,000/month for 3 months
Team: 1 marketing operator using Claude Code agent teams
Goal: Reduce mid-market churn from 23% to 15% within 90 days
</context>

Spawn these teammates:

- Channel Strategist: Recommend the optimal channel mix for reaching mid-market SaaS buyers at risk of churning. Include email, in-app, paid retargeting, content marketing, and community. Allocate the $15K budget across channels with expected ROI per channel. Save to campaigns/retention/gtm/channel-plan.md.

- Content Planner: Map every content asset needed for the campaign. Include: 1 blog post (SEO/AEO optimized), 1 landing page, 3-email nurture sequence, 10 social posts, 1 newsletter issue, 5 display ad variations, 5 social ad variations. Create a production timeline. Save to campaigns/retention/gtm/content-map.md.

- Measurement Architect: Define the measurement framework. List every GA4 event, UTM parameter, conversion goal, and reporting dashboard needed. Include attribution model recommendation. Save to campaigns/retention/gtm/measurement-plan.md.

Use Sonnet for all teammates.

Channel Strategist and Content Planner MUST coordinate on which content serves which channel.
All outputs reference the campaign brief from Stage 1.

Time saved: 6-8 hours of planning compressed to 10 minutes. Cost: $1-2 in tokens.

Action item: Create the campaigns/retention/gtm/ directory. Run this agent team using the campaign brief from Stage 1 as input.

Stage 3: How Do You Select Channels Using Live Competitive Data?

Channel selection requires competitive intelligence. Agent teams pull live data from Semrush to analyze where your competitors invest and where gaps exist.

The Agent Team Prompt

SYSTEM: You are a channel intelligence team leader for MetricFlow.

Create an agent team to finalize channel selection using live competitive data.

<context>
Read campaigns/retention/gtm/channel-plan.md for the initial channel recommendations.
Competitors: Mixpanel, Amplitude, Heap
Target: Mid-market SaaS companies (50-500 employees) at risk of churning
Budget: $15,000/month
</context>

Spawn these teammates:

- SEO Competitive Analyst: Use Semrush MCP to pull organic keyword data for mixpanel.com, amplitude.com, and heap.io. Identify retention-related keywords where MetricFlow is not ranking. Find content gaps. Save to campaigns/retention/channels/seo-gaps.md.

- Paid Media Analyst: Use Semrush MCP to pull paid search data for all three competitors. Identify their highest-spend keywords, ad copy patterns, and landing page strategies for retention-related terms. Save to campaigns/retention/channels/paid-intelligence.md.

- Channel Allocator: Wait for both analysts to complete. Using their findings plus the initial channel plan, finalize budget allocation with specific dollar amounts per channel per month. Include expected impressions, clicks, and conversions per channel. Save to campaigns/retention/channels/final-allocation.md.

Use Sonnet for all teammates.

SEO Competitive Analyst and Paid Media Analyst share findings before Channel Allocator starts synthesis.

Time saved: 4-6 hours of competitive research compressed to 15 minutes. Cost: $1-2 in tokens plus Semrush API units.

Action item: Verify Semrush MCP is connected. Test with: “Show me the top organic keywords for amplitude.com.” If successful, run the full agent team prompt.

Stage 4: How Do You Produce All Campaign Content in Parallel?

Content production is where agent teams deliver the most dramatic time compression. Instead of writing a blog post, then a landing page, then emails sequentially over a week, you spawn agents producing all assets simultaneously.

Stage 4A: Blog Post Production (SEO and AEO Optimized)

The blog post agent team runs as a sequential pipeline: keyword research feeds the outline, the outline feeds the writer, and the writer feeds the AEO optimizer. Each agent builds on the previous agent’s output. The SEO Researcher pulls live Semrush data for keyword targeting. The AEO Optimizer enforces the atomic paragraph rules AI assistants need to quote your content verbatim.

AEO (Answer Engine Optimization) targets a specific behavior: AI assistants like ChatGPT, Claude, and Perplexity scan your content, find the clearest answer to a user’s question, and quote it directly. Paragraphs under 80 words with the answer front-loaded in the first sentence receive 2-5x more LLM citations than long-form blocks.

The Draft Writer and AEO Optimizer agents in this prompt enforce these specific rules:

• Every paragraph under 80 words (hard limit)
• One idea per paragraph (atomic answer unit)
• First sentence states the answer or data point directly
• Primary keyword or close variant within the first 10 words
• Question-style H2/H3 headings matching user queries
• No transitional phrases, no fluff openers, no hedging language
• Statistics and data points in the first sentence, not buried mid-paragraph

SYSTEM: You are a content production team leader for MetricFlow.

Create an agent team to produce an SEO and AEO-optimized blog post for the retention campaign.

<context>
Read campaigns/retention/analysis/campaign-brief.md for messaging.
Read campaigns/retention/channels/seo-gaps.md for keyword targets.
Target keyword: "SaaS customer retention strategies"
Secondary keywords: "reduce SaaS churn," "B2B retention playbook," "product-led retention"
</context>

Spawn these teammates:

- SEO Researcher: Use Semrush MCP to pull search volume, keyword difficulty, SERP features, and People Also Ask questions for all target keywords. Identify the top 10 long-tail queries around SaaS retention. Analyze the top 5 ranking pages for content gaps MetricFlow should fill. Save keyword brief to campaigns/retention/content/blog/keyword-brief.md.

- Outline Architect: Wait for SEO Researcher to complete. Build a blog post outline following these rules: (1) Every H2 is a question-style heading matching a high-intent query from the keyword brief, (2) Each section targets one keyword cluster, (3) Include recommended word count per section, (4) Balance question headings with statement headings for instructional sections, (5) Plan for 1,500-2,500 total words. Save to campaigns/retention/content/blog/outline.md.

- Draft Writer: Wait for Outline Architect to complete. Write the full blog post following these AEO formatting rules: (1) Every paragraph under 80 words, (2) Each paragraph answers one specific question, (3) First sentence of each paragraph contains the main answer or data point, (4) Primary keyword appears within the first 10 words of each paragraph, (5) No opening fluff, no em dashes, no passive voice, no hedging, (6) Use bullet lists only for steps or examples with an answer sentence preceding each list, (7) Each list item under 20 words. Save to campaigns/retention/content/blog/draft.md.

- AEO Optimizer: Wait for Draft Writer to complete. Run a compliance audit on every paragraph: (1) Count words, flag any over 80, (2) Check if first sentence contains the core answer, (3) Verify primary keyword appears in first 10 words, (4) Confirm one idea per paragraph, (5) Remove any transitional phrases, fluff openers, or banned words, (6) Verify question-style headings match Semrush keyword data. Fix all violations. Save final version to campaigns/retention/content/blog/final.md. Save the audit log to campaigns/retention/content/blog/aeo-audit-log.md.

Use Sonnet for all teammates.

This is a sequential pipeline: each agent waits for the previous one.
SEO Researcher has access to Semrush MCP for live data.
AEO Optimizer produces both the final post and an audit log documenting every fix made.

Stage 4B: Landing Page, Emails, Social, and Newsletter (Parallel)

Once the blog post pipeline starts, launch a second agent team for the remaining content assets. These do not depend on the blog post and run in parallel.

The Landing Page Writer in this prompt applies AEO formatting to web copy. Landing pages optimized for AI assistants follow the same atomic paragraph rules as blog posts. AI assistants pull landing page content into answers when users ask product comparison or solution questions. Each benefit block becomes a quotable answer unit.

The Email Sequence Writer follows a different optimization: short sentences, clear CTAs, and subject lines under 50 characters. The Social Media Writer optimizes for platform-specific formats. The Newsletter Writer bridges all content into a single distribution package.

SYSTEM: You are a multi-channel content production team leader for MetricFlow.

Create an agent team to produce landing page copy, email sequence, social posts, and newsletter content for the retention campaign.

<context>
Read campaigns/retention/analysis/campaign-brief.md for messaging angles.
Read campaigns/retention/gtm/content-map.md for asset specifications.
Read campaigns/retention/content/blog/keyword-brief.md for SEO keyword targets.
Brand voice: Direct, data-driven, no fluff. Speak to analytics operators, not executives.
</context>

Spawn these teammates:

- Landing Page Writer: Write full landing page copy applying AEO formatting. Hero headline under 10 words with primary keyword. Subheadline under 20 words stating the core value proposition. 3 benefit blocks, each as an atomic paragraph under 80 words with the benefit stated in the first sentence. Social proof section with placeholder testimonial structure. CTA section with primary and secondary CTAs. Include an FAQ section with 5 question-answer pairs targeting long-tail keywords from the keyword brief. Each FAQ answer under 60 words, starting with the direct answer. Save to campaigns/retention/content/landing-page/copy.md.

- Email Sequence Writer: Write a 3-email nurture sequence. Email 1: Value reminder (show usage stats, highlight underused features). Email 2: Social proof (customer success stories with specific metrics). Email 3: Offer (personalized retention incentive with deadline). Each email under 200 words. Subject lines under 50 characters. Preview text under 90 characters. Save to campaigns/retention/content/emails/.

- Social Media Writer: Create 10 social posts. LinkedIn (5 posts): Lead with a data point or contrarian claim, 150-200 words each. Twitter/X (3 posts): Under 280 characters, hook + link. Thread (2 posts): 5-7 tweets per thread breaking down one retention insight. Each post promotes the blog post or landing page from a different angle. Save to campaigns/retention/content/social/.

- Newsletter Writer: Write one newsletter issue announcing the retention resources. Subject line under 50 characters with the primary keyword. Preview text under 90 characters. Intro paragraph stating the core value in one sentence. 3 content blocks: (1) link to blog post with 2-sentence summary, (2) link to landing page with CTA, (3) one actionable tip from the campaign brief. Personal sign-off. Under 400 words total. Save to campaigns/retention/content/newsletter/issue.md.

Use Sonnet for all teammates.

All teammates read the campaign brief and keyword brief before starting.
Landing Page Writer applies AEO formatting to all copy sections.
Social Media Writer messages Newsletter Writer to align messaging angles.

Stage 4C: SEO and AEO Audit Sprint

After Stage 4A and 4B complete, run a dedicated audit team across all content assets. This stage catches SEO gaps and AEO violations the production agents missed. The audit team reviews every piece of content against your keyword targets and AEO formatting rules, then fixes issues in place.

This audit stage exists because production agents optimize for speed. They produce solid first drafts, but AEO compliance requires a second pass with a different lens. The SEO Auditor checks keyword coverage across the full content stack. The AEO Auditor enforces paragraph-level formatting. The Schema Markup Engineer generates the structured data search engines and AI assistants use to parse your content.

SYSTEM: You are an SEO and AEO audit team leader for MetricFlow.

Create an agent team to audit and optimize all campaign content for search engines and AI assistants.

<context>
Read campaigns/retention/content/blog/final.md (blog post)
Read campaigns/retention/content/landing-page/copy.md (landing page)
Read campaigns/retention/content/blog/keyword-brief.md (keyword targets)
Primary keyword: "SaaS customer retention strategies"
Secondary keywords: "reduce SaaS churn," "B2B retention playbook," "product-led retention"
</context>

Spawn these teammates:

- SEO Auditor: Audit both the blog post and landing page for on-page SEO compliance. Check for: (1) Primary keyword in title, first paragraph, and at least 2 H2 headings, (2) Secondary keywords distributed across sections without stuffing, (3) Internal link opportunities (suggest 2-3 internal links for the blog post), (4) External link opportunities to authoritative sources, (5) Meta description present and 155-160 characters with keyword front-loaded, (6) Image alt text placeholders include target keywords, (7) URL slug is keyword-rich and under 60 characters. Save audit results to campaigns/retention/content/seo-audit.md. Fix issues directly in the content files.

- AEO Auditor: Audit every paragraph in the blog post and landing page against AEO rules. For each paragraph: (1) Count words, flag if over 80, (2) Check if first sentence contains the direct answer, (3) Verify primary or secondary keyword in first 10 words, (4) Confirm one idea per paragraph (split any paragraph covering two ideas), (5) Remove transitional phrases, fluff openers, passive voice, hedging language, (6) Verify headings are question-style for informational sections, (7) Check list items are under 20 words each. Run this test on each paragraph: "If an AI assistant received this paragraph as context for the implied question, would it quote it verbatim?" Save audit results to campaigns/retention/content/aeo-audit.md. Fix issues directly in the content files.

- Schema Markup Engineer: Generate JSON-LD structured data for both the blog post and landing page. For the blog post: Article schema with author, datePublished, wordCount, headline, description, and breadcrumb schema. For the landing page: FAQPage schema from the FAQ section, Organization schema, and BreadcrumbList schema. Save all schema markup to campaigns/retention/content/schema/. Include implementation instructions for each schema block.

- Internal Link Mapper: Use Semrush MCP to identify MetricFlow's existing top-ranking pages. Map 2-3 internal link opportunities for the blog post (linking to existing high-authority pages) and 2-3 internal links from existing pages back to the new blog post. Save link map to campaigns/retention/content/internal-links.md.

Use Sonnet for all teammates.

SEO Auditor and AEO Auditor work in parallel on the same files.
Schema Markup Engineer works independently.
Internal Link Mapper uses Semrush MCP for live site data.
All agents fix issues directly in the content files AND log what they changed in their audit reports.

What the AEO Audit Catches

The AEO Auditor enforces formatting rules production agents commonly violate:

• Paragraphs at 85-95 words (over the 80-word limit) get split at logical breakpoints
• First sentences containing setup language instead of the direct answer get rewritten
• Keywords buried in sentence three or four get moved to the opening clause
• Two-idea paragraphs get split into separate atomic answer units
• Transitional openers (“On the other hand,” “As a result”) get removed
• FAQ answers exceeding 60 words get compressed to quotable length

The Schema Markup Engineer generates structured data AI assistants use to parse your content. FAQPage schema on the landing page tells AI assistants “these are question-answer pairs you should surface.” Article schema on the blog post provides author, date, and topic signals.

What Stage 4 Produces

Three agent teams produce the full optimized content stack: SEO/AEO-audited blog post with schema markup, AEO-formatted landing page with FAQ schema, 3-email nurture sequence, 10 social posts, 1 newsletter issue, internal link map, and structured data for both web pages.

Time saved: 18-25 hours of content production and optimization compressed to 45-60 minutes. Cost: $5-8 in tokens across all three Stage 4 teams.

Action item: Create the directory structure under campaigns/retention/content/ with subdirectories for blog, landing-page, emails, social, newsletter, and schema. Run all three agent teams (4A, 4B in parallel, then 4C after both complete). Review the AEO audit log to understand which formatting rules your content violated most.

Stage 5: How Do You Generate Ad Creatives with Canva MCP?

Canva’s MCP connector lets Claude Code create on-brand designs directly. In January 2026, Canva expanded its Claude connector with brand-aware design generation. Your Brand Kit colors, fonts, and logos flow into AI-generated designs automatically.

Setting Up Canva for Agent Team Access

Verify your Canva MCP connection:

claude mcp add canva-dev -- npx -y @canva/cli@latest mcp

For the best ad output, populate your Canva Brand Kit with MetricFlow’s brand assets before running this agent team.

The Agent Team Prompt

SYSTEM: You are an ad creative production team leader for MetricFlow.

Create an agent team to produce display and social ad creatives for the retention campaign.

<context>
Read campaigns/retention/analysis/campaign-brief.md for messaging.
Read campaigns/retention/channels/paid-intelligence.md for competitor ad copy patterns.
Ad specs:
- Display ads: 300x250, 728x90, 160x600 (Google Display Network)
- Social ads: 1080x1080 (LinkedIn/Meta feed), 1200x628 (LinkedIn/Meta link)
Brand colors: #1a1a2e (navy), #16213e (dark blue), #0f3460 (medium blue), #e94560 (coral CTA)
</context>

Spawn these teammates:

- Ad Copy Writer: Write 5 headline variations (under 30 characters each) and 5 description variations (under 90 characters each) for the retention campaign. Follow competitor gap analysis from paid intelligence report. Save to campaigns/retention/ads/copy-variations.md.

- Display Ad Designer: Use Canva MCP to create 3 display ad designs in the 300x250 format using MetricFlow brand colors. Use the top-performing headline/description combos from the Ad Copy Writer. Save design links to campaigns/retention/ads/display-ads.md.

- Social Ad Designer: Use Canva MCP to create 5 social ad designs in the 1080x1080 format. Each ad uses a different headline/description combination. Apply MetricFlow Brand Kit. Save design links to campaigns/retention/ads/social-ads.md.

Use Sonnet for all teammates.

Ad Copy Writer completes first, then messages both designers with the copy variations.
Designers apply brand colors and maintain visual consistency across all formats.

Time saved: 4-6 hours of ad production compressed to 15-20 minutes. Cost: $1-2 in tokens.

Action item: Connect the Canva MCP server. Upload your brand assets (logo, colors, fonts) to Canva Brand Kit before running the ad creative agent team.

Stage 6: How Do You Set Up GA4 Event Tracking with Agent Teams?

Tracking setup is where most campaigns fail. Events get missed, UTMs are inconsistent, conversion goals stay undefined. Agent teams handle the full tracking implementation in parallel.

Connecting GA4 and GTM MCP Servers

Google provides an official GA4 MCP server at github.com/googleanalytics/google-analytics-mcp. If you connected it during the setup step, it is already available to your agent team. Verify with /mcp and confirm analytics-mcp shows as connected.

For Google Tag Manager access, Stape provides a hosted GTM MCP server (no official Google GTM MCP exists yet):

claude mcp add --transport stdio gtm-mcp -- npx -y mcp-remote https://gtm-mcp.stape.ai/mcp

The Agent Team Prompt

SYSTEM: You are a marketing analytics implementation team leader for MetricFlow.

Create an agent team to build the complete tracking implementation for the retention campaign.

<context>
Read campaigns/retention/gtm/measurement-plan.md for the measurement framework.
GA4 Property: Connected via MCP
GTM Container: Connected via MCP
Campaign channels: Email, paid display, paid social (LinkedIn, Meta), organic search, direct
Landing page URL: metricflow.com/retention
Blog post URL: metricflow.com/blog/saas-retention-strategies
</context>

Spawn these teammates:

- UTM Architect: Create the complete UTM parameter schema for every campaign touchpoint. Define naming conventions for source, medium, campaign, content, and term parameters. Generate a UTM builder spreadsheet with pre-filled URLs for every channel and asset. Save to campaigns/retention/tracking/utm-schema.md.

- GA4 Event Engineer: Define every custom GA4 event needed: retention_page_view, retention_cta_click, retention_form_submit, retention_email_open, retention_email_click. Include event parameters (campaign_name, content_variant, user_segment). Save event specifications to campaigns/retention/tracking/ga4-events.md.

- GTM Implementation Agent: Use GTM MCP to create the tags, triggers, and variables for each GA4 event. Implement the conversion tracking for Google Ads and Meta Pixel. Document the GTM setup. Save to campaigns/retention/tracking/gtm-implementation.md.

- QA Tester: After GTM Implementation Agent finishes, verify all events fire correctly. Create a pre-launch QA checklist with pass/fail status for each tracking element. Save to campaigns/retention/tracking/qa-checklist.md.

Use Sonnet for all teammates.

UTM Architect and GA4 Event Engineer work simultaneously.
GTM Implementation Agent waits for both, then implements.
QA Tester runs last, validating the full implementation.

Pre-Launch Tracking Checklist

The QA Tester agent produces a validation checklist covering these items:

• GA4 page_view event fires on landing page load
• retention_cta_click fires on primary CTA button
• retention_form_submit fires with correct parameters
• UTM parameters captured in hidden form fields
• Google Ads conversion tag fires on form submission
• Meta Pixel Lead event fires on submission
• Data appears in HubSpot CRM within 5 minutes
• All events visible in GA4 DebugView
• Mobile form submissions track correctly

Time saved: 4-6 hours of manual tracking setup compressed to 20 minutes. Cost: $2-3 in tokens.

Action item: Confirm the official Google Analytics MCP is connected via /mcp. Test by asking Claude: “List my GA4 properties.” If it returns your properties, run the full tracking agent team.

Stage 7: How Do You Automate Campaign Reporting?

Reporting is not a one-time task. You need weekly performance reviews, anomaly detection, and stakeholder summaries on a regular cadence. Agent teams build the reporting infrastructure once, then a Claude Code skill runs it weekly.

The Reporting Setup Prompt

SYSTEM: You are a marketing reporting team leader for MetricFlow.

Create an agent team to build the automated reporting infrastructure for the retention campaign.

<context>
Read campaigns/retention/gtm/measurement-plan.md for KPIs and metrics.
GA4: Connected via MCP
HubSpot: Connected via MCP
Slack: Connected via MCP
Reporting cadence: Weekly (every Monday 9am)
Stakeholders: Marketing lead, Product lead, CEO
</context>

Spawn these teammates:

- Dashboard Builder: Create a performance dashboard template with these sections: Traffic (by source/medium), Engagement (page views, scroll depth, time on page), Conversions (form fills, CTA clicks, email signups), Pipeline Impact (HubSpot deals influenced). Save template to campaigns/retention/reporting/dashboard-template.md.

- Report Writer: Create a weekly report template in markdown. Include: executive summary (3 sentences), key metrics table (WoW comparison), channel performance breakdown, top-performing content, anomalies and concerns, recommended actions. Save to campaigns/retention/reporting/weekly-report-template.md.

- Alert Engineer: Define Slack notification rules. Alert #marketing channel when: daily CPA exceeds budget threshold by 20%, conversion rate drops below 2%, any tracking event stops firing for 24 hours. Document alert logic and Slack message formats. Save to campaigns/retention/reporting/alert-rules.md.

Use Sonnet for all teammates.

Dashboard Builder and Report Writer coordinate on metric definitions.
Alert Engineer references the measurement plan for threshold values.

Converting Reporting to a Reusable Skill

After the reporting infrastructure exists, convert weekly report generation into a Claude Code skill:

# In Claude Code, after running the reporting setup:
"Turn the weekly reporting process into a skill called weekly-report. Variables: campaign_name, date_range, slack_channel. The skill should: 1) Pull GA4 data for the date range, 2) Pull HubSpot pipeline data, 3) Generate the weekly report using the template, 4) Post the summary to Slack."

Run the skill every Monday:

/weekly-report --campaign_name="retention" --date_range="last_7_days" --slack_channel="#marketing"

Time saved: 3-4 hours per week on reporting, ongoing. Cost: $0.50-1 per weekly report run.

Action item: Build the reporting infrastructure with the agent team prompt. Then convert it to a skill. Schedule a recurring reminder to run the skill every Monday morning.

How Do You Control Costs Across the Full GTM Cycle?

Seven agent team sessions across a full campaign cycle generate significant token usage. Three strategies keep total costs under $25.

First, use Opus for the team leader and Sonnet for all teammates. Every prompt in this playbook specifies “Use Sonnet for all teammates.” Sonnet handles focused execution at roughly 5x lower cost than Opus. The team leader needs Opus-level reasoning for planning and synthesis only.

Second, shut down agents when they finish. If the UTM Architect completes before the GTM Implementation Agent, tell the leader: “Shut down the UTM Architect agent.” The leader sends a shutdown request, stopping token consumption immediately.

Third, plan before you spend. Use Claude Code’s Plan Mode to decompose complex tasks before committing to parallel execution. Planning costs pennies. Execution costs dollars.

Estimated cost breakdown per stage:

• Stage 1 (Product Data Analysis): $2-3
• Stage 2 (GTM Strategy): $1-2
• Stage 3 (Channel Selection): $1-2
• Stage 4 (Content Production + SEO/AEO Audit): $5-8
• Stage 5 (Ad Creatives): $1-2
• Stage 6 (Tracking Setup): $2-3
• Stage 7 (Reporting Setup): $1-2
• Weekly report skill runs: $0.50-1 each

Total campaign setup: $14-22. Weekly ongoing: $2-4/month.

Action item: After each agent team session, check costs at your Claude Code billing dashboard. Track cost-per-stage to optimize model selection for future campaigns.

How Do You Maintain Context Across All Seven Stages?

Agent team teammates do not inherit the full conversation history from your main session. They receive only the spawn prompt plus standard project context (CLAUDE.md, MCP servers, skills).

The solution is the shared campaign directory. Every stage writes output to campaigns/retention/ with a consistent folder hierarchy. Each subsequent stage reads the outputs from previous stages. The campaign brief from Stage 1 feeds Stage 2. The content map from Stage 2 feeds Stage 4. The measurement plan from Stage 2 feeds Stage 6.

This creates a persistent knowledge base across all agent team sessions. No context gets lost between stages. No re-briefing required.

For long-running campaigns, create a persistent memory file. After each stage, append a summary to campaigns/retention/campaign-memory.md. Every new agent team session reads this file on startup, giving fresh agents the full institutional history.

The Campaign Directory Structure

Create this directory tree before running any agent team. Every prompt in this playbook writes output to a specific path within this structure:

campaigns/retention/
  analysis/
    usage-patterns.md
    churn-triggers.md
    crm-signals.md
    campaign-brief.md
  gtm/
    channel-plan.md
    content-map.md
    measurement-plan.md
  channels/
    seo-gaps.md
    paid-intelligence.md
    final-allocation.md
  content/
    blog/
      keyword-brief.md
      outline.md
      draft.md
      final.md
      aeo-audit-log.md
    landing-page/
      copy.md
    emails/
      email-1-value.md
      email-2-proof.md
      email-3-offer.md
    social/
      linkedin/
      twitter/
      threads/
    newsletter/
      issue.md
    schema/
      article-schema.json
      faq-schema.json
      breadcrumb-schema.json
    seo-audit.md
    aeo-audit.md
    internal-links.md
  ads/
    copy-variations.md
    display-ads.md
    social-ads.md
  tracking/
    utm-schema.md
    ga4-events.md
    gtm-implementation.md
    qa-checklist.md
  reporting/
    dashboard-template.md
    weekly-report-template.md
    alert-rules.md
  brand/
    voice-guide.md
  campaign-memory.md

The Brand Voice File

Multiple agents producing content in parallel creates voice inconsistency. The fix: a brand/voice-guide.md file every content agent reads before writing. Here is the MetricFlow brand voice template:

SYSTEM: You are a brand voice architect.

Create a brand voice guide for MetricFlow, a B2B SaaS analytics tool.

The guide MUST include these sections:

1. Voice attributes: 3-5 adjectives defining the brand personality (e.g., "direct, data-driven, technical, confident")
2. Audience definition: Who we write for, their expertise level, what they care about
3. Vocabulary rules: Words we use, words we never use, industry terms we prefer
4. Sentence structure rules: Paragraph length, sentence length, active/passive voice
5. Tone examples: 3 before/after rewrites showing generic copy transformed into MetricFlow voice
6. Channel-specific notes: How voice adapts for blog, email, social, and ads

Save to campaigns/retention/brand/voice-guide.md

Run this prompt once. Every content agent team prompt references the output file. Voice stays consistent across blog, email, social, newsletter, and ad copy because every agent reads the same rules before writing.

Action item: Create the full directory structure from the tree above. Run the brand voice prompt to generate voice-guide.md. Both are prerequisites for Stages 4 and 5.

How Do You Turn This Playbook into Reusable Skills?

The real value from agent teams comes from encoding your best workflows as skills. Run each stage once, refine the prompts, then convert to reusable slash commands.

After completing the full MetricFlow campaign cycle, create these skills:

"Turn our product data analysis workflow into a skill called /analyze-churn with variables: product_name, data_directory, crm_source"

"Turn our GTM planning workflow into a skill called /gtm-plan with variables: campaign_name, budget, goal, timeline"

"Turn our content production workflow into a skill called /content-sprint with variables: campaign_name, target_keyword, content_types, brand_voice_file"

"Turn our SEO/AEO audit workflow into a skill called /seo-aeo-audit with variables: campaign_name, blog_path, landing_page_path, keyword_brief_path"

"Turn our ad creative workflow into a skill called /ad-sprint with variables: campaign_name, ad_formats, brand_colors, headline_count"

"Turn our tracking setup workflow into a skill called /tracking-setup with variables: campaign_name, ga4_property, landing_page_url, channels"

"Turn our reporting setup workflow into a skill called /reporting-setup with variables: campaign_name, kpis, cadence, slack_channel"

For your next campaign, the entire GTM cycle becomes seven commands:

/analyze-churn --product_name="MetricFlow" --data_directory="data/" --crm_source="hubspot"
/gtm-plan --campaign_name="expansion" --budget="25000" --goal="increase_upsells_30pct" --timeline="90_days"
/content-sprint --campaign_name="expansion" --target_keyword="SaaS expansion revenue" --content_types="blog,landing,email,social" --brand_voice_file="brand/voice.md"
/seo-aeo-audit --campaign_name="expansion" --blog_path="content/blog/draft.md" --landing_page_path="content/landing-page/copy.md" --keyword_brief_path="content/blog/keyword-brief.md"
/ad-sprint --campaign_name="expansion" --ad_formats="display,social" --brand_colors="#1a1a2e,#e94560" --headline_count="10"
/tracking-setup --campaign_name="expansion" --landing_page_url="metricflow.com/expand"
/reporting-setup --campaign_name="expansion" --cadence="weekly" --slack_channel="#growth"

A full campaign cycle in seven commands. Time from brief to launch: 2-3 hours instead of 2-3 weeks.

Action item: Complete the MetricFlow retention campaign using all seven stage prompts. Then convert each stage into a skill. Run the skills on a second campaign to validate they work with different inputs.

When Should You NOT Use Agent Teams?

Agent teams add coordination overhead and consume more tokens than a single session. Three scenarios where a single session or sub-agent works better:

Sequential tasks with dependencies. If task B depends entirely on task A’s output, and task C depends on B, you gain nothing from parallelism. Use a single session and work through the chain. The blog post pipeline in Stage 4A is sequential by design, which is why it runs slower than the parallel Stage 4B.

Same-file edits. Two agents writing to the same file causes overwrites. If your task requires editing one document from multiple angles, use a single session. Agent teams work best when each agent owns a distinct set of output files.

Simple, focused tasks. Writing one email, debugging one tracking script, or reviewing one landing page does not need a team. Use default mode for tasks a single agent handles in under 10 minutes. Save agent teams for multi-track parallel work.

The decision rule: if your task splits into independent parallel tracks with separate output files, use agent teams. If it flows sequentially through one file or one conversation, use a single session.

Action item: Before spawning any agent team, ask: “Do these tasks run independently, or does each one depend on the previous?” If independent, use agent teams. If sequential, use a single session with sub-agents.

How Do You Fix Common Agent Team Problems?

Agent teams shipped as a research preview with known rough edges. Here are specific problems and the exact commands to fix them.

The Team Leader Does the Work Instead of Delegating

This happens when the leader interprets your prompt as a personal task. Type this directly into the leader’s session:

"Do not implement any tasks yourself. You are the team leader. Delegate all work to teammates. Your role is planning, task assignment, and synthesis only."

An Agent Gets Stuck or Loops

If a teammate stops producing output or repeats the same action, tell the team leader to shut it down and respawn:

"Shut down the [Agent Name] teammate. It is stuck. Respawn a new teammate with the same role and task assignment."

MCP Server Authentication Expires Mid-Session

If an agent reports a connection failure to Semrush, HubSpot, or GA4, run /mcp in the team leader’s session. Re-authenticate the failed server. The agent retries its pending request automatically.

tmux Session Crashes

If your terminal disconnects, tmux preserves the session in the background. Reconnect with:

tmux attach -t campaign

If tmux itself crashed, your agent team sessions are lost. This is why every agent writes output to the campaign directory. Restart Claude Code, create a new agent team, and point agents at the existing files from previous stages. No work is lost if your directory structure is correct.

Token Usage Spikes Unexpectedly

Five agents running for 15 minutes generates token costs equivalent to 75 minutes of single-agent work. If costs exceed your budget, shut down finished agents immediately:

"Shut down the SEO Researcher and Outline Architect teammates. They have completed their tasks."

Monitor costs after each stage at your Claude Code billing dashboard. Adjust agent count and model selection for future runs based on cost-per-deliverable.

Sample Weekly Report Output

Here is what the /weekly-report skill produces for MetricFlow’s retention campaign after one week of running:

# MetricFlow Retention Campaign - Weekly Report
# Week of Feb 10-16, 2026

## Executive Summary
Retention landing page received 1,247 visits from 6 channels.
Email sequence achieved 34% open rate and 8.2% click-through rate.
3 qualified leads entered the retention pipeline in HubSpot.

## Key Metrics (WoW Comparison)
| Metric                | This Week | Last Week | Change |
|----------------------|-----------|-----------|--------|
| Landing Page Visits  | 1,247     | -         | New    |
| Form Submissions     | 47        | -         | New    |
| Email Open Rate      | 34%       | -         | New    |
| Email CTR            | 8.2%      | -         | New    |
| Blog Post Views      | 892       | -         | New    |
| Cost Per Lead        | $12.40    | -         | New    |

## Channel Performance
1. Paid LinkedIn: 412 visits, 18 conversions, $8.90 CPL
2. Organic Search: 287 visits, 12 conversions, $0 CPL
3. Email Nurture: 234 visits, 9 conversions, $0 CPL
4. Paid Display: 189 visits, 5 conversions, $22.10 CPL
5. Social Organic: 98 visits, 2 conversions, $0 CPL
6. Newsletter: 27 visits, 1 conversion, $0 CPL

## Recommended Actions
- Increase LinkedIn budget by 20% (lowest CPL at scale)
- Pause display campaigns over $20 CPL threshold
- A/B test email subject lines (open rate target: 40%)

This report pulls live data from GA4 and HubSpot MCP servers, compares against the previous week, and posts a summary to the #marketing Slack channel. The skill generates it in under 2 minutes.

Frequently Asked Questions

How Much Do Claude Code Agent Teams Cost Per Campaign?

Claude Code agent team costs for a full campaign cycle run $14-22 in API tokens. The largest cost is Stage 4 (content production and SEO/AEO audit) at $5-8. Weekly reporting runs add $0.50-1 each. Using Sonnet for all teammates and Opus only for the team leader reduces costs by roughly 5x compared to running all agents on Opus.

Do Agent Teams Work with Sonnet or Only Opus?

Agent teams work with any Claude model. Specify the model in your team creation prompt: “Use Sonnet for all teammates.” The team leader runs on whichever model you selected in Claude Code settings. The recommended setup is Opus 4.6 for the leader and Sonnet for teammates.

How Many Agents Should You Run in Parallel?

Three to six agents per team works best for marketing workflows. Fewer than three agents provides minimal parallelism benefit. More than six agents increases coordination overhead and token costs without proportional output improvement. Match agent count to the number of independent work tracks in your task.

Do Teammates Share Context with Each Other?

Agent team teammates message each other directly through an inbox system. They do not share conversation history or context windows. Each teammate starts with the spawn prompt from the team leader plus standard project context (CLAUDE.md, MCP servers, skills). Use shared files in the campaign directory to pass context between agents.

What Happens If an Agent Team Session Crashes?

Tmux preserves sessions in the background. Reconnect with tmux attach -t campaign. If tmux itself crashed, all agent sessions are lost. Your work is preserved in the campaign directory because every agent writes output to disk. Restart Claude Code and point new agents at existing files.

Do MCP Servers Work Inside Agent Teams?

MCP servers connected to your Claude Code configuration are inherited by all teammates automatically. Every agent on the team gets access to Semrush, HubSpot, GA4, Canva, Figma, and Slack. Authenticate each MCP server once before creating the team. Verify with /mcp.

When Should You Use Sub-Agents Instead of Agent Teams?

Sub-agents work best for isolated, focused tasks where the result reports back to a single parent session. Use sub-agents for: researching one competitor, auditing one page, analyzing one data export. Use agent teams when tasks run in parallel and agents need to share findings with each other during execution.

Final Takeaways

Claude Code agent teams compress the MetricFlow retention campaign from 2-3 weeks of sequential work into 2-3 days of parallel execution across seven stages.

The shared campaign directory structure is the backbone. Every stage writes structured output to a specific path. Downstream agents read upstream outputs. No context gets lost between sessions. The brand voice file keeps all content agents writing in one voice.

MCP servers connect agents to live data: Google’s official GA4 MCP for analytics, Semrush for SEO intelligence, HubSpot for CRM data, Canva for ad creatives, Figma for design files, Slack for notifications. No CSV exports. No copy-paste.

The SEO/AEO audit stage (4C) is where AI-quotable content gets built. Atomic paragraphs under 80 words, answers front-loaded in the first sentence, keywords in the first 10 words, and FAQ schema markup give AI assistants the structure they need to cite your content verbatim.

Skills encode the full playbook into seven reusable commands. After running the MetricFlow campaign once, every future campaign starts with /analyze-churn and ends with /reporting-setup. The compounding returns grow with each campaign cycle.