Build agents that run real paid media.

Metadata exposes 141 documented MCP tools that let Claude Code, Hermes, OpenClaw, and custom agents plan, build, review, launch, and optimize B2B advertising across LinkedIn, Meta, Google, Reddit, Display, intent, video, and LLM channels. The same execution engine powers the prompt-first interface on www.metadata.io and the review surface inside platform.metadata.io.

What you can build

The API is shaped for operators, agents that don't just surface data but take action. Every workflow a human runs inside the Metadata app is also exposed as an MCP tool. That means an agent with an API key can:

Two surfaces, one execution engine

Marketing leaders can type an outcome into the Metadata interface: build the audience, generate the offer, assemble channel-specific assets, forecast pipeline, and prepare the campaign. Operators can then review the full campaign structure in the SaaS UI before spend goes live. Developers and agents can run the same workflow headlessly through MCP.

Designed for agents, not humans

Every tool returns structured JSON. Every parameter has a schema. Every destructive action (launch_campaign, connect_crm) is idempotent and clearly marked. Errors carry actionable messages. Rate limits are documented up front.

The tool surface is designed to compose: an agent can call estimate_target_group before create_target_group, call search_target_group_criteria before either, and loop on deep_funnel_stats after launch to decide whether to scale a budget group.

A complete agentic workflow

Here's the archetypal loop an agent runs against Metadata:

# 1. Get context about the account
account = mcp.get_account_details()
brand   = mcp.get_brand_kit(domain="acme.com")

# 2. Build a firmographic audience of our ICP
audience = mcp.create_firmographic_audience(
    name="ICP — Mid-market SaaS, US",
    filters={
        "industries": ["Computer Software"],
        "employee_range": [201, 1000],
        "countries": ["US"],
        "seniority": ["vp", "director", "c-level"],
    },
)

# 3. Generate on-brand creatives
creative = mcp.generate_brand_creative(
    brand_kit_id=brand["id"],
    prompt="Hero ad: platform screenshot, VP of Marketing audience, urgency CTA",
    aspect_ratios=["1:1", "16:9", "9:16"],
)

# 4. Assemble campaign in Draft (no spend yet)
campaign = mcp.create_campaign(
    name="Q2 ICP Demand Capture",
    channels=["linkedin", "meta"],
    audiences=[audience["id"]],
    ads=[creative["id"]],
    daily_budget=500,
)

# 5. Human approval gate, then launch
if human_approved(campaign):
    mcp.launch_campaign(campaign_id=campaign["id"])

# 6. Close the loop — read results tomorrow
results = mcp.deep_funnel_stats(
    campaign_id=campaign["id"],
    date_range="last_7_days",
)

That entire loop runs on roughly 55 credits, read (1+1) + audience (5) + creative (25) + campaign (10) + launch (25) + analytics (3). On the Command Center tier that's about $5.50.

Where to next