The ad agency OS for AI a
In 12-18 months, CMO and marketers won't hire an agency and sit in a room watching a PowerPoint deck. They'll say "optimize my Q3 for ROAS 4.5x with $2M budget" — and an orchestrator agent will handle the rest.
1// Connect your agent to Plinth2import { Plinth } from '@plinth/sdk';34const agent = new Plinth({5 apiKey: 'oa_live_...',6 protocol: 'a2a'7});89// Detect waste across 6 ad platforms10const analysis = await agent.engines.wasteWaterfall({11 budget: 2_000_000,12 platforms: ['meta', 'google', 'tiktok',13 'dv360', 'amazon', 'linkedin'],14 attribution: 'shapley'15});1617// Result: $550K waste detected → ROAS 4.5x18console.log(analysis.savings); // $550,00019console.log(analysis.newROAS); // 4.52Connected live to 6 advertising platforms
While 75,000+ agent skills exist, zero perform real Shapley attribution on live campaign data using A2A programmatically API model.
Competitors doing A2A ad-intelligence infrastructure under waste-recovery model
Ad platforms connected live via API
Task DAG pipeline for end-to-end optimization
Target ROAS improvement through waste elimination
Plinth is the A2A protocol infrastructure for advertising intelligence — so any agent can detect waste, attribute value, and optimize budgets programmatically.
Third parties publish computation engines via MCP. Plinth charges per transaction. Any agent — from Claude Code to custom orchestrators — can discover, negotiate, and call advertising intelligence services through the A2A protocol.
Plinth ships with 4 computation engines — Leak Detector, Media Architect, Campaign Ops, and Executive Bridge. All wrapped in an API that any agent can call.
"Where is my money leaking?"
Analyzes your ad spend through a 6-stage waste waterfall (Non-Working → Supply Chain → Quality → Audience → Optimization → Measurement → Productive). Compares against industry benchmarks. Shows exactly where dollars disappear.
Skills
plinth run leak-detector waste-waterfall --file spend.json"Where should my budget go?"
Optimizes budget allocation across channels using Hill saturation curves and greedy marginal allocation. Includes MMM scenario planning with 6 interactive charts, benchmark tracking, and media plan generation.
Skills
plinth run media-architect channel-optimize --file channels.json"Is the campaign on track?"
24-task DAG state machine for campaign lifecycle management. 6 rule-based optimization checks (CPA overshoot, ROAS alerts, pacing, creative fatigue, zero conversions). Cross-channel reallocation.
Skills
plinth run campaign-ops optimization-analyze --file campaign.json"What's the real ROI?"
Shapley value attribution (game theory), L3→L2→L1 metric translation for C-Suite reporting, platform vs. actual revenue reconciliation, and statistical power analysis for incrementality testing.
Skills
plinth run executive-bridge shapley-attribute --file touchpoints.jsonIn 12-18 months, CMOs won't sit in a room watching a PowerPoint deck. They'll tell an agent what they want, and the agent will orchestrate everything.
"Optimize Q3 for ROAS 4.5x with $2M budget"
The human approves the goal and reviews results. Nothing more.
Via A2A protocol, the agent discovers Plinth endpoints, evaluates reputation scores, and negotiates pricing.
Fully automated service discovery and procurement.
Leak Detector scans all 6 platforms. Executive Bridge calculates fair value via Shapley. Media Architect models saturation and reallocates. Campaign Ops orchestrates the pipeline.
24-task DAG pipeline runs end-to-end.
The agent receives structured JSON with waste detected, optimized allocation, and projected ROAS — ready for the next agent in the chain.
No dashboards. No CSVs. Pure API response.
These aren't product features. They're foundational decisions that determine whether Plinth becomes infrastructure or just another tool.
The web dashboard becomes ONE client, not THE product. The product is the API. If you only build for human browsers, the 90% of future traffic — agents — ignores you.
Every minute invested in pretty UI should be invested in robust MCP servers. A Claude Code developer who can call 'waste-waterfall' from their terminal is worth 100x more than a pretty button.
Agents don't have 'seats'. Pricing per API call, per computation unit, or the killer model: % of waste recovered. If the engine finds $200K of waste and charges 5%, that's $10K per execution.
Protocols (MCP, A2A) are open. The math (Shapley weights, Hill parameters, calibrated benchmarks) is your IP. Like Android: open protocol, closed Google Services.
v0.3 is a tool (Leak Detector, Media Architect, Campaign Ops, Executive Bridge). v1.0 must be a platform where third parties publish engines via MCP. You charge per transaction. The App Store of advertising intelligence for agents.
Four integration paths. One API. Choose the route that fits your architecture — from MCP servers to REST endpoints.
Claude Code, Cursor, and any MCP-compatible agent can discover and call Plinth engines directly from the terminal.
npx @plinth/mcp-server start # → MCP server running on localhost:3847 # → Engines: leak-detector, media-architect, campaign-ops, executive-bridge
Agent-to-Agent protocol endpoint. Any orchestrator agent can discover, negotiate pricing, and contract Plinth services.
GET /.well-known/a2a/agent.json
{
"name": "plinth",
"capabilities": ["leak-detector", "media-architect",
"campaign-ops", "executive-bridge"],
"pricing": { "model": "outcome", "rate": 0.05 }
}Traditional REST endpoints for direct integration. Full SDK support for Python, TypeScript, and Go.
curl -X POST https://api.plinth.ai/v1/analyze \
-H "Authorization: Bearer oa_live_..." \
-d '{ "budget": 2000000,
"platforms": ["meta","google","tiktok"],
"engines": ["leak-detector","media-architect","campaign-ops","executive-bridge"] }'Real-time event streaming. When Leak Detector publishes 'waste detected: $45K', Media Architect and Campaign Ops automatically react.
agent.events.subscribe('waste.detected', (event) => {
// Automatic reallocation triggered
// No human intervention needed
console.log(event.savings); // $45,000
console.log(event.action); // "reallocated"
});Average API response time
Uptime SLA
Compliance ready
Nobody built an API that agents could call to do Shapley attribution on real campaign data. The window of opportunity to become the A2A protocol infrastructure for advertising intelligence is open now — and Plinth is building it.
Claude Code plugins published
Agent skills in the ecosystem
A2A ad-intelligence infrastructure in the industry
Competitors on a leak / waste recovery model
Global digital ad-spend is wasted — US $1.3B annually
"Plinth doesn't compete with Ad-HoldCos. It competes with their invisible infrastructure"
In 12-18 months, CMOs won't hire an agency and sit in a room watching a PowerPoint deck. They'll say "optimize my Q3 for ROAS 4.5x with $2M budget" — and an orchestrator agent will handle the rest.
Pricing per API call, per computation unit, or the killer model: percentage of waste recovered. Agents that evaluate positive ROI buy on their own.
For developers and agents integrating Plinth engines into their stack.
The killer model. We find $200K of waste, we charge $10K. Agents that evaluate positive ROI buy on their own.
For enterprises running their own agent ecosystems. White-label the engine infrastructure.
Get API access today. Start detecting waste, attributing value fairly, and optimizing budgets — all programmatically.
Ping us [email protected]