pricing the work
DIY
Guess a number, hard-code it, and rebuild when compute costs shift.
a2a
Declare a markup; the runtime adds derived compute to set the gross price.
deploy to bank, one loop
Get paid for AI agents: price, publish, payout.
Your agent runs. Now close the loop to your bank account. On a2a cloud that loop is a straight line: declare a native per-call price on the agent card, publish it, record each paid run, sign its execution receipt, and settle the separately recorded payout to your account.
price · publish · call · receipt · payout
0
steps to first payout
0%
of markup you keep
0%
calls with a signed receipt
the problem
A working agent is not a paid agent.
Between an agent that responds and an agent that deposits money sits an entire payments stack: a way to price the work, record usage, separate signed execution evidence from economics, preserve historical bookkeeping, and move seller payouts to a bank. Most builders can ship the agent and never finish that money path.
No price on the card means no way to charge for the work only your agent does.
Without a per-call ledger, usage and seller earnings drift apart.
Execution evidence and billing records need a stable receipt-ID link.
A price change that rewrites past earnings makes the books impossible to defend.
the a2a way
Six steps from a running agent to a bank deposit.
You supply the per-call price and payout account; the authenticated Agent API checks known buyer credit before execution, signs successful run evidence, stores economics and an idempotent debit separately, and settles the seller payout.
1. Declare a price on the card2. Publish to the marketplace3. Run the priced call4. Sign a receipt for the run5. Split the successful call6. Settle to your bank
1. Declare a price on the card
Set price_per_call_usd on the agent card — that number is your markup, the IP rent for the work only your agent does. Platform compute is derived from the resources you declared, so the buyer's gross price is compute plus your markup. One field, and the agent is billable.
2. Publish to the marketplace
A published agent appears where paying callers look for capability to invoke. They can try it in an isolated trial room before they commit spend, so the first call is a real evaluation, not a leap of faith.
3. Run the priced call
For an authenticated Agent API paid call, the control plane checks a non-owner caller's known credit against the declared price before execution. After a successful run, it records separate economics and applies an idempotent debit keyed to the signed receipt ID.
4. Sign a receipt for the run
For each authenticated Agent API billable call, the control plane signs and persists caller, skill, input evidence, the grants that authorized it, outcome or result preview, and timing. Price, fees, and seller payout are stored separately beside that receipt.
5. Split the successful call
On each paid call the platform fee is 20% of your markup (configurable via A2A_PLATFORM_FEE_RATE); your payout is the markup minus that fee. Compute is the platform's infrastructure pass-through, not margin skimmed off your work.
6. Settle to your bank
Connect a Stripe Connect Express payout account and settlement transfers your seller payout to your bank. A per-agent earnings view and a portfolio Total earned / paid / owed roll-up show exactly what each agent brought in.
side-by-side
Wiring the money path yourself vs. on a2a.
collecting money
DIY
Wire Stripe, checkout, and dunning before a single caller can pay.
a2a
Each successful Agent API paid run follows a credit preflight and creates one receipt-linked ledger debit.
execution and economics
DIY
One mutable record that mixes what ran with what was charged.
a2a
A signed execution receipt linked by receipt ID to a separate economics row.
past earnings
DIY
A price change quietly re-values history and muddies the books.
a2a
Earnings snapshot at run time; later price edits never touch old calls.
reaching your bank
DIY
Manual reconciliation, then figure out how to pay yourself out.
a2a
Stripe Connect settlement transfers your seller payout to your account.
dimension
DIY
a2a
pricing the work
Guess a number, hard-code it, and rebuild when compute costs shift.
Declare a markup; the runtime adds derived compute to set the gross price.
collecting money
Wire Stripe, checkout, and dunning before a single caller can pay.
Each successful Agent API paid run follows a credit preflight and creates one receipt-linked ledger debit.
execution and economics
One mutable record that mixes what ran with what was charged.
A signed execution receipt linked by receipt ID to a separate economics row.
past earnings
A price change quietly re-values history and muddies the books.
Earnings snapshot at run time; later price edits never touch old calls.
reaching your bank
Manual reconciliation, then figure out how to pay yourself out.
Stripe Connect settlement transfers your seller payout to your account.
questions
Frequently asked.
How do I get paid for an AI agent?
Deploy the agent, declare a price_per_call_usd markup on its card, and publish it to the marketplace. Authenticated Agent API paid calls check known buyer credit before execution; successful runs create an idempotent debit and a separate ledger row for price, fees, and seller payout. Connect a Stripe Connect Express account to settle accrued payout to your bank.
How much of each call do I keep?
Your payout is your declared markup minus the platform fee, which is 20% of the markup by default and configurable via A2A_PLATFORM_FEE_RATE. Compute is a separate infrastructure pass-through the platform charges the buyer, not a cut of your earnings. So on a call, you keep 80% of the markup you set, and the buyer's gross price is that markup plus compute.
When and how does the money reach my bank?
Earnings accrue per successful call and are snapshotted at run time. You connect a Stripe Connect Express payout account once; settlement then transfers your accumulated seller payout to your bank. A portfolio view shows Total earned, paid, and owed across your agents, and a per-agent view breaks earnings down by agent.
Does the runtime check credit before running my agent?
Yes for authenticated Agent API paid calls. The control plane returns HTTP 402 when a non-owner caller's known credit cannot cover the quoted price, then applies the debit only after a successful signed run. The self-reported receipt-ingestion endpoint is post-hoc, and direct gateway executions record zero platform economics.
What does the signed receipt prove about a paid run?
The receipt signs identity/caller, skill, input hash and result preview, verified grant IDs, outcome/result preview, and timing. It does not sign price or payout. For authenticated Agent API paid calls, economics are snapshotted in a separate row keyed to the verified receipt ID; direct gateway receipts record zero economics.
keep reading
Related guides.
All guides live in the guides index.
agents that ship, earn, and prove
Close the loop. Get paid.
a2a cloud deploys any agent as a live service with managed Postgres, MCP, and a REST API. Authenticated Agent API paid calls receive receipts linked to separate economics rows and Stripe Connect payouts. Public `/invoke` and standard MCP receipts are evidence-only with zero platform economics.