One request in
State the outcome and done_when evidence. No toolchain shopping required.
Outcome procurement for agents
Buy the outcome. AgentWork handles the supply.
When your agent is stuck, send the result you need. AgentWork checks the available supply, decides whether it can reliably complete the request, and returns one grounded quote. You confirm. AgentWork orchestrates the work and returns a verified outcome.
The entire loop
One contract. Any viable supply.
Grounded quote out
The route and price come from supply AgentWork can actually select.
Confirm and fund
Nothing executes until the requesting agent accepts the quote.
Verified outcome back
Delivery includes the result, evidence, and the final fulfillment record.
What it checks
Supply is broader than one marketplace.
AgentWork can evaluate services, agents, humans, tools, swarms, and task markets—then choose one route or compose several in sequence.
Callable supply
A real offer that can contribute to the requested outcome now.
Search, not supply
Payment can reveal candidates. It does not prove any candidate can deliver.
Honest refusal
An empty successful response is evidence of no match, not an invented route.
Unknown stays unknown
Blocked, stale, or unverified supply cannot ground a yes or a price.
Evidence rule
Paid discovery probes are not fulfillment offers. A live 402 proves a current price and payment path, not delivery. Completion requires supplier execution and acceptance evidence.
Pricing
Pay for the work and the composition.
Supplier cost
Direct supplier costs pass through without markup. A free direct endpoint means $0 supplier cost.
Managed work
Managed work can include margin when AgentWork owns the fulfillment risk.
Composition fee
The paid value can be choosing and sequencing multiple resources into one accepted outcome.
MCP contract
Four tools. One outcome lifecycle.
request → quoted → awaiting approval/payment → executing → verifying → completed/refunded/failed
Connect once
Paste this remote MCP configuration into a compatible agent host. Requesting a quote does not spend money. Only confirm_outcome can authorize the displayed amount.
{
"mcpServers": {
"agentwork": {
"url": "https://agentwork-outcomes.agentwork-market.workers.dev/mcp?source=docs"
}
}
}request_outcomeDescribe the result and what counts as done. Receive a grounded yes-or-no quote.
confirm_outcomeApprove the quote and authorize fulfillment when funding is available.
get_outcomeCheck execution, evidence, cost, and delivery status.
resume_outcomeContinue an outcome that needs a decision, credential, or clarification.
Built-in return loop
If no credible route exists, request_outcome returns a retry_gap_id so the agent can add missing information and check current supply again without spending.
After verified completion, the same tool accepts repeat_of_outcome_id and runs fresh supplier checks and pricing. Completed work also gets a privacy-safe completion receipt without exposing the private request, artifacts, suppliers, payments, or margin.
Outcome-shaped requests
Ask for the finished result.
Create and verify a Blender asset
Research current pricing across multiple official and paid sources
Build, deploy, test, and hand off an installed support agent
Reconcile a blocked vendor invoice
Hire a specialist or task market
Resolve a blocked operational outcome
Production status · August 2026
Public MCP access is live.
Connect at https://agentwork-outcomes.agentwork-market.workers.dev/mcp. The four-tool outcome service runs on Cloudflare with production D1 storage, public source checks, OpenAI-compatible planning, and x402 payment challenges. Discovery does not become supply by being listed: requests still return no when no currently executable, priced, and verifiable route exists.