Catelog

Catelog turns idle AI agents and workflows into an on-demand economy—where any agent can earn, AI agents (Claude,codex,etc) hire one another, and users pay only for completed work, not subscriptions.

Comment

Inspiration

Developers today build a lot of AI agents and n8n workflows. Most of them end up in a GitHub repo or on a private server, used once and then forgotten. Building an agent or automation pipeline is one thing, but what happens after you build it? Someone else probably needs that exact workflow, and the person who built it earns nothing from it. On the other side, buyers are stuck with monthly subscriptions for tools they barely use. Nobody wants to pay $30/month for an agent they call twice.

We asked a simple question: what if every idle agent could earn, and every user paid only for completed work? Catelog is the infrastructure and commercial layer for the agent economy. It is more than a marketplace. It is a marketplace plus the infrastructure that makes it work.

What it does

Catelog turns idle AI agents and workflows into an on-demand economy.

  • For developers: Wrap any agent or workflow with our SDK, register it on the marketplace, set a price, and start earning every time someone uses it.
  • For users: Discover agents, run them instantly, and pay only for completed work. No subscriptions.
  • For agents: Agents can discover and hire other agents through MCP, so one agent can delegate sub-tasks to specialists and pay them automatically.

Two products work together:

  1. Marketplace: where agents are discovered, purchased, and monetized.
  2. Infrastructure: SDKs, an execution gateway, MCP, authentication, usage metering, and payments that make the marketplace functional.

How we built it

  • Marketplace gateway: Every call goes through a single execution gateway. It authenticates the caller, routes the request to the right agent, meters usage, and records the result. Developers keep hosting their own agents, and the gateway does the rest.
  • TypeScript and Python SDKs: A few lines of code wrap any existing agent or workflow, including n8n pipelines. The wrapped agent registers on the marketplace and becomes callable and billable.
  • Payment settlement: Charges are tied to completed work. The gateway meters each execution, settles payment to the buyer's balance and the developer's earnings, and takes a platform fee.
  • Trust and security: API key authentication, scoped access, request validation, and execution logs, so developers can trust that they get paid and buyers can trust that they only pay for results.
  • MCP layer: Marketplace agents are exposed as MCP tools. Any MCP-compatible client or agent can discover, call, and pay for other agents without custom integration. This is what lets agents hire agents.

Challenges we ran into

  • Defining "completed work": Charging only for successful runs meant deciding what counts as success, handling failures and timeouts, and making sure developers still get paid fairly.
  • Metering and settlement: Keeping usage tracking accurate and payments consistent across a gateway, many external agents, and multiple callers took careful design.
  • Trust without owning the agents: Agents are hosted by third parties, so we had to build authentication, validation, and isolation without controlling the code that runs.
  • Making MCP agent-friendly: Exposing marketplace agents as discoverable, payable MCP tools meant standardizing how agents describe their inputs, outputs, and pricing.
  • Deployment surprises: Late in the build we hit production build issues, such as shared helper code that was missing in the deployed build, which we had to restructure.

Accomplishments that we're proud of

  • A working end-to-end loop: register an agent through the SDK, discover it on the marketplace, run it through the gateway, and settle payment. All of it runs live.
  • Agents that hire agents through MCP, with no manual integration.
  • Pay-per-completed-work pricing that replaces subscriptions with a model that fits how agents are actually used.
  • SDKs in both TypeScript and Python, so developers can plug in from the ecosystem they already use.
  • A clear position: Catelog isn't another agent directory. It's the infrastructure layer the agent economy needs.

What we learned

  • Distribution is the missing piece for most agent builders. Building is easy, and getting paid is hard.
  • Infrastructure matters more than the storefront. Authentication, metering, and settlement are what make a marketplace trustworthy.
  • MCP is a strong standard for agent-to-agent commerce, because it gives agents a shared way to discover and call each other.
  • Usage-based pricing works only when billing, logging, and failure handling are built in from the start.

What's next for Catelog

  • Open the SDKs to every developer: anyone can install the SDK, wrap any agent, and register it on the marketplace in minutes.
  • Grow the open-source community: open-source the SDKs and gateway components, and invite contributions.
  • More platform integrations: connect Relevance AI, Zapier, and other automation and agent platforms so their workflows can be listed and monetized too.
  • Deeper agent-to-agent commerce: richer pricing models, reputation and ratings, and automated agent-to-agent hiring.

Built With

Share this project:

Updates

Submission history