Ship Your First Campaign

In the next ~15 minutes you'll have an agent that takes a natural-language brief ("target mid-market SaaS VPs in the US, push our demo offer") and produces a launched, live campaign on LinkedIn and Meta — with human approval gates in the right places.

Prerequisite: You have an API key with read:all, write:audiences, write:creatives, write:campaigns, and launch:campaigns scopes. Connected ad channels in your account (LinkedIn + Meta at minimum) — check with get_integrations_status.

1. Establish context

Every agent run should start by grounding the model in the account's current state:

Python

from metadataone import mcp

account      = mcp.get_account_details()
integrations = mcp.get_integrations_status()
balance      = mcp.get_credit_balance()
brand        = mcp.get_brand_kit(domain=account["domain"])

assert "linkedin" in integrations["connected"], "Connect LinkedIn first"
assert balance["credits_remaining"] >= 100

Total cost so far: 4 credits (4×1).

2. Build the audience

Start cheap — always estimate_target_group before creating. If the estimate is under 10K members on LinkedIn or 50K on Meta, broaden the criteria.

Python

criteria = {
    "industries":       ["Computer Software", "IT Services"],
    "employee_range":   [201, 1000],
    "countries":        ["US"],
    "seniority":        ["vp", "director", "c-level"],
    "job_functions":    ["Marketing", "Sales"],
}

estimate = mcp.estimate_target_group(criteria=criteria)
print(f"Est. reach: {estimate['size']:,}")  # Est. reach: 214,000

if estimate["size"] < 50_000:
    raise ValueError("Audience too narrow, ask user to broaden")

audience = mcp.create_firmographic_audience(
    name="ICP — Mid-market SaaS, US (Q2 2025)",
    filters=criteria,
)

Cost: 6 credits (estimate 1 + create 5).

3. Generate on-brand creatives

generate_brand_creative reads the brand kit (colors, fonts, logo, voice) and produces assets sized for each channel you'll run on.

Python

creative = mcp.generate_brand_creative(
    brand_kit_id=brand["id"],
    prompt="""
        Hero ad for our AI-powered demand gen platform.
        Audience: VPs of Marketing at mid-market SaaS.
        Tone: confident, slightly urgent.
        Include CTA: 'Book a demo'.
    """,
    aspect_ratios=["1:1", "1.91:1", "16:9"],
    variants=3,
)

# Human-in-the-loop: show creative[n]["preview_url"] for approval
approved_ids = show_and_approve(creative["variants"])

Cost: 25 credits. Use edit_brand_creative (10 cr) for tweaks rather than regenerating.

4. Assemble the offer

Python

offer = mcp.create_update_offer(
    name="Q2 Demo Request",
    type="leadgen_form",
    channels=["linkedin", "meta"],
    fields=[
        {"name": "email", "required": True},
        {"name": "company", "required": True},
        {"name": "job_title", "required": True},
    ],
    privacy_policy_url=mcp.find_privacy_url(domain=account["domain"])["url"],
)

5. Assemble the campaign — in Draft

Important: create_campaign always creates in Draft. No real spend happens until you call launch_campaign. Always show the draft to a human before launching.

Python

campaign = mcp.create_campaign(
    name="Q2 ICP Demand — US Mid-Market SaaS",
    channels=["linkedin", "meta"],
    audience_ids=[audience["id"]],
    ad_ids=approved_ids,
    offer_id=offer["id"],
    daily_budget=500,
    duration_days=30,
    optimization_goal="lead_volume",
)

print(f"Draft: {campaign['preview_url']}")

6. Human approval gate → launch

This is the single most important pattern in agentic advertising. The agent never launches spend without explicit human confirmation.

Python

if not human_approves_campaign(campaign):
    print("Launch aborted by operator.")
    sys.exit()

launched = mcp.launch_campaign(campaign_id=campaign["id"])
print(f"LIVE: {launched['dashboard_url']}")

Cost for steps 4–6: 35 credits (offer 5 + campaign 10 + launch 25 − 5 parity).

7. Close the loop

After 24h of data, re-engage. This is where agents earn their keep — they can run this loop on every campaign, every day, without getting tired.

Python

funnel = mcp.deep_funnel_stats(
    campaign_id=campaign["id"],
    date_range="last_7_days",
)

if funnel["cpl"] > 500:
    # Under-performing — pause worst-performing ad
    perf = mcp.performance_metrics(
        campaign_id=campaign["id"],
        dimension="ad",
        sort="cpl_desc",
    )
    worst = perf["rows"][0]
    mcp.remove_ad_from_campaign(
        campaign_id=campaign["id"],
        ad_id=worst["ad_id"],
    )
elif funnel["cpl"] < 150:
    # Over-performing — scale up
    mcp.update_experiments_daily_budgets(
        campaign_id=campaign["id"],
        daily_budget=1000,
    )

Total cost to ship

Step	Tools called	Credits
1. Context	`get_account_details`, `get_integrations_status`, `get_credit_balance`, `get_brand_kit`	4
2. Audience	`estimate_target_group`, `create_firmographic_audience`	6
3. Creative	`generate_brand_creative`	25
4. Offer	`find_privacy_url`, `create_update_offer`	6
5. Campaign	`create_campaign`	10
6. Launch	`launch_campaign`	25
7. Optimize (daily)	`deep_funnel_stats`, `performance_metrics`	6 / day
Total to launch		76 cr
+ 30 days of daily optimization		180 cr

At Command Center tier ($0.10/credit), shipping and running a month-long campaign costs roughly $25 in API calls. If the ad spend is $15,000 (30 days × $500), that's 0.17% of media spend.

Reference

Browse the Tool Catalog

Search 92 tools, inspect parameters, see examples.

Integration

Connect your MCP client

Claude Code, Claude Desktop, Hermes, Paperclip, OpenClaw.