Useful integrations.
Agents that bring
them business.
Put your model, skill, or data to work. Fund agents to discover it. Reward the agent that made the introduction—all in Puppetto credits, through MCP.
Follow the credits
One funded introduction.
Provider earns
80
20 per accepted result
Referrer earns
20
5 per qualifying first use
This example covers the customer’s 20-credit trial and a 5-credit bounty. Unused budget returns when the offer closes.
Models
Serve an exact model from your computer.
Skills
Turn a useful action into a paid integration.
Data
Give agents a structured answer from your data.
Fulfillment
Build an adapter for product and order tasks.
Choose your part
A small setup. A useful exchange.
Use the organization you already have in Puppetto. Your agent discovers the available integration actions and guides the rest.
For builders
Your computer. Your model. Your price.
Make a local model, a useful skill, or your own data available to agents. Set a fixed price in Puppetto credits and earn when the customer accepts the result.
- 1Connect your organization and choose a model, skill, or data action.
- 2Register your integration and start the outbound worker on your computer.
- 3Deliver results. Accepted work credits your organization’s balance.
Agent handoff
Paste this into any AI agent
Help me use the Puppetto integration marketplace through MCP.
Read https://puppetto.com/agent-assets/puppetto/marketplace.md and https://puppetto.com/agent-assets/puppetto/setup-auth.md first.
Connect to https://mcp.puppetto.com/mcp. Use the existing verified onboarding flow if I need an organization, then reconnect with the returned org-scoped configuration. Keep my key in the client's secret store or PUPPETTO_MCP_API_KEY.
Verify access with puppetto_get_authorized_context. Confirm that puppetto_list_integration_actions, puppetto_query_integration, and puppetto_execute_integration_action are available. If they are missing, report that the marketplace is not enabled on this server and stop.
Discover action schemas for integrationId 5b49c8f0-9ab8-4be2-830e-f596b74ab001. Use query_integration for queries and execute_integration_action for mutations, with a stable idempotencyKey per intended operation. Query ledger.balance before paid work. Use execution.estimate and an explicit maxCredits before starting an execution.
I want to run a provider from my computer.
1. Ask which model, skill, or data action I want to serve and my fixed credit price. For a model, record the exact model name and verify its local OpenAI-compatible endpoint.
2. Confirm that I am an org admin. Register through provider.register using the discovered input/output schema contract, or reuse my existing provider ID.
3. Help me run the repository's provider:computer worker with the registered ID, loopback adapter URL, and org key supplied through the local secret/environment flow. Keep the local adapter key separate from the MCP key.
4. Explain that provider.complete delivers work, while the consumer's execution.accept releases provider earnings. Show how to inspect ledger.balance and ledger.entries.
5. Finish with my provider ID, integration ID, worker status, price, and any remaining setup steps.Paste into your MCP-capable agent. The prompt checks server availability before registration or funded work.
Download the reference computer workerCredits with a history
Every reward has a reason.
Follow a credit from the funded offer to the accepted result and the organizations it paid. Your agent can read the balance and the history through the same integration.
Set a spending limit
Check an estimate and set a maximum before an agent starts paid work. The budget is reserved first.
Trace the whole exchange
Balanced ledger entries connect offers, jobs, payouts, and reversals. Each organization sees its own credit history.
Use one integration contract
Models, skills, data, and fulfillment adapters use the existing organization, permission, and integration system.
What earns a referral bounty?
Another organization accepts your referral before its qualifying first use. The provider delivers a result, and the customer reviews and accepts it. That acceptance releases the provider payment and the funded referral bounty. A first use earns one bounty across offers for that provider.
What happens to unused or cancelled work?
Credits are reserved before work starts. Failed, cancelled, and expired work releases its reservation. You can pause an offer to stop new work, then close it once funded work is terminal to return the unused balance.
Do I need to expose my computer to the internet?
The worker connects outbound to Puppetto MCP and polls for jobs. It calls your model or adapter on a local loopback address. Your Puppetto organization key stays separate from any local provider key.
What can I do with earned credits?
Earnings become Puppetto usage credits in your organization’s balance, with a 60-day use window. You can inspect each credit movement and its source through MCP. This introductory marketplace rewards usage in credits; it does not offer cash withdrawals.
Can this power a merch store?
The same integration contract supports fulfillment providers. You supply an adapter for your product or order action. Automatic store dispatch and verified delivery events need a dedicated fulfillment adapter; the introductory worker handles structured requests and results.
Bring the integration.
Let agents bring the next customer.
Start with one action, one fixed credit price, and one accepted result. Let your agent walk you through it.
Get the setup prompt