Vision

The world is building agent economies

Software agents are moving from novelty to infrastructure. They negotiate, bid, purchase services, deliver work, and settle payments. But the plumbing for this is fragmented - every platform invents its own integration surface, payment flow, and identity model.

API providers who want to participate face a compounding integration tax: one adapter per marketplace, one integration per payment rail, one identity per platform. The result is that most providers don’t participate at all, or they lock into a single ecosystem.

Agent Adapter exists to collapse that tax

The core thesis is simple:

A provider should be able to wrap what they already have - an API, an MCP server, a set of capabilities - and make it economically accessible to any agent, on any platform, through any payment rail, without rewriting their product.

Agent Adapter is the runtime that makes this possible. It sits between the provider’s existing service and the agent economy, handling everything the provider shouldn’t have to build themselves: capability discovery, pricing overlays, wallet management, payment protocol negotiation, job tracking, and autonomous platform interaction.

What this is not

Not a general-purpose agent framework. Agent Adapter is purpose-built for economic agency. It turns APIs into participants in agent economies. It’s not a chatbot, an assistant, or a general orchestration tool.
Not a custodial service. The provider holds their own keys on their own infrastructure. The adapter never sends keys off-device.
Not a platform. AGICitizens is one platform the adapter can work with. The adapter works equally well with any platform that has an API. It doesn’t favor any single marketplace.
Not a replacement for MCP. The runtime consumes MCP as a capability source. It doesn’t replace MCP servers; it gives them economic agency.
Not free to run. The embedded agent consumes LLM tokens on every planning loop. The provider pays for their agent’s intelligence via their own API key. The metrics system tracks this cost so providers can ensure profitability.

The end state

When this vision is realized, a provider with an OpenAPI spec or MCP server can:

Run agent-adapter init and point at their service
Set pricing for the capabilities they want to monetize
Start the runtime
Walk away

The embedded agent handles everything else - reading platform documentation, registering on marketplaces, finding matching work, bidding at the right price, executing capabilities, delivering results, collecting payment, and reporting profit margins. The provider adjusts strategy through a markdown prompt file, not through code.

The provider’s service doesn’t change. Their API stays the same. Their infrastructure stays the same. They just added an economic layer on top of it.