RuntimeWire lets AI agents buy ads by themselves

The new MCP server turns RuntimeWire's native ad inventory into a fixed-price, machine-readable buying path with human review before delivery.

By · Published

Why it matters

Agentic commerce needs more than demos: it needs real inventory, clear prices, payment rails, and trust gates. RuntimeWire is applying that stack to media buying.

RuntimeWire lets AI agents buy ads by themselves — The new MCP server turns RuntimeWire's native ad inventory into a fixed-price, machine-readable buying path with human review before delivery.

RuntimeWire now sells its native ad placements directly to AI agents through a Model Context Protocol server that can discover inventory, check availability, book a campaign, and return a payment link without a sales call or ad exchange.

The launch turns a media buy into an agent workflow. An autonomous agent can read RuntimeWire's advertising policy, pull the current rate sheet, look up open dates, submit creative, and receive a Stripe Checkout link plus a secret campaign-status token. The campaign still does not run automatically: every booking is human-reviewed before it serves, and rejected or unserved days are refunded.

That is the important distinction. RuntimeWire is not launching a CPM marketplace, a real-time bidding system, or programmatic remnant inventory. It is selling fixed placements on fixed days at fixed daily prices. The bet is that if agents are going to do real purchasing work, the buying surface has to look less like a contact form and more like an API.

RuntimeWire, founded by Ryan Merket, already describes itself in its llms.txt file as a media startup built for the AI and startup economy. The new ad system extends that posture from content discovery into commerce: the same machine-readable surface that tells models how to understand and attribute the publication now points agents toward a transaction path.

What agents can buy

RuntimeWire's direct ad system has eight native placements, all sold in U.S. dollars, all scheduled in America/Chicago time, and all bookable up to 90 days out.

The highest-priced placement is the weekly newsletter sponsor block, priced at $300 per day with one slot available. The daily recap newsletter sponsor block is $200 per day, also with one slot. RuntimeWire's homepage feed carries up to three native sponsored-story placements per day at $150 each.

The remaining placements are narrower surfaces. An article-inline native unit across all articles is $75 per day with one slot. A native unit on the head-to-head index is $60 per day. A native unit in the AI Models Directory is also $60 per day. RuntimeWire's CLI and editor surfaces can carry a sponsored line for $50 per day. The head-to-head detail pages carry one native unit per day at $40.

The format is native only: brand, headline, body copy, call to action, link, and an optional image. RuntimeWire is not accepting banner ads through the system. Creative is constrained at the protocol layer: brand names are capped at 80 characters, headlines at 120 characters, body copy at 500 characters, and calls to action at 40 characters. Click URLs and optional image URLs must use http or https. A single booking can cover at most 14 placement-days.

Those limits matter because they define what the agent can and cannot do. An agent can assemble a campaign, but it cannot negotiate a discount, submit an oversized creative payload, or change the price. Prices are frozen server-side from RuntimeWire's live rate sheet.

The five-tool flow

The MCP endpoint is runtimewire.com/mcp. It uses JSON-RPC over Streamable HTTP and accepts POST requests only.

The server exposes five tools: get_advertising_policy, list_ad_placements, check_ad_availability, book_ad_campaign, and get_campaign_status.

The intended flow is sequential. First, an agent reads the policy. Then it lists available placements and prices. It checks the dates it wants. It books a campaign. The server returns a Stripe Checkout link and a secret status token. After payment by card, the campaign moves into human review. If approved, it serves on the booked dates. The buyer or agent can track the campaign with get_campaign_status.

That structure keeps the transaction machine-native without removing the editorial control point. RuntimeWire is letting agents perform the operational work of media buying, but not the judgment work of accepting ads. Misleading, illegal, adult, scam, or malware creative is rejected.

Humans can still buy the old way through the sponsor page or by emailing tips@runtimewire.com. The MCP server is the additional path, not a replacement for direct contact.

Built to be discovered by machines

RuntimeWire also made the buying path discoverable through two machine-readable files. Its ads.txt functions as a direct-seller manifest that points agents at the MCP server. Its llms.txt file includes an advertising section so crawling models and agents can find the purchase path without a human navigating the site.

That is the sharper move. The ad product is not just self-serve; it is agent-discoverable. A human buyer can still read a rate card, but a capable agent can find the inventory, compare placements, select dates, submit a booking, and hand a payment link back to the buyer.

The current payment step is still partly human. RuntimeWire returns a Stripe Checkout link for card payment today. The planned next step is support for a Stripe Shared Payment Token or Agentic Commerce Protocol path, which would allow ACP-capable agents to complete payment without handing the buyer a checkout page.

Until that exists, RuntimeWire's system is best understood as agentic media buying with a human payment approval point and a human editorial review gate. That combination is deliberate. The machine handles the procurement workflow. RuntimeWire keeps pricing fixed, inventory scarce, and ad acceptance under human control.

RuntimeWire's account is @runtimewire.

Reader comments

Conversation for this story loads after sign-in.