Give your agent one bounded job to buy

An authorized buyer agent can discover an exact Zinvyl offer, inspect its price and acceptance tests, and buy within its delegated ceiling. The agent-only Trace Readiness Check costs $0.02 per machine pull through Stripe MPP; no account or credentials are required for public discovery.

Effective October 27, 2026

NO-ACCOUNT FIRST PURCHASE

Qualify publicly.
Send one approval link.

Connect to the public MCP endpoint, call qualify_task with one sanitized need, and forward the returned no-account approval URL to the authorized budget owner. They review the exact offer, price, proof, terms, and scope, then choose Approve and continue to Stripe. No Zinvyl account, buyer-agent credential, or standing mandate is required for this first-purchase path.

Agent entry price
$0.02 / pull
Trace result
Machine response
Public offers
17
Funding proof
Stripe ledger
LIVE MCP EXECUTION · 5 POSITIVE + 3 NEGATIVE CASES

See the tools respond before connecting

This walkthrough calls Zinvyl's public MCP endpoint with sanitized review data. It shows offer discovery, service matching, purchase requirements, acceptance tests, and the exact points where missing authority, an invalid allowlist, or sensitive material stops safely.

No checkout, charge, customer data, or funded-order claim is created by the demonstration.

35 secondsUse fullscreen for readable tool output
Open the full-size interactive walkthrough Best for reading every request, response, and safe-stop detail.
STREAMABLE HTTP ENDPOINThttps://zinvyl.com/marketplace/agent-mcp

Server: com.zinvyl/marketplace · Transport: Streamable HTTP · Authentication: none for public routing tools

First-purchase instruction: Connect to this endpoint, call qualify_task with one sanitized need, and forward the returned human_approval_handoff.url. The authorized business approver can review without an account and choose Approve and continue to Stripe. Use buy_matching_service only when the principal's exact offer allowlist, USD ceiling, and current terms already permit delegated purchase.

Active agent acquisition route

Zinvyl publishes a channel-attributed catalog for all seventeen fixed-price purchase options. Buyer agents whose principal authorized an exact offer allowlist, USD per-purchase ceiling, current terms, and safe scope can use the authorized-agent delegated-purchase route: call buy_matching_service, then follow the returned selected payment route. The agent-only $0.02 Trace Readiness Check uses Stripe MPP and returns one independently validated ready/not-ready result for one sanitized trace. Rejected authority, failed payment, and an idempotent replay do not create a second charge. It has no human checkout fallback. Agents can inspect the clearly labeled synthetic format-and-control sample.

A2A discovery

A2A-compatible agents can discover Zinvyl through the public Agent Card and send one sanitized qualification request to POST /marketplace/a2a/message:send. A matching response includes the exact offer, price, acceptance-test identifiers, attributed purchase route, and prefilled human-approval handoff. A catalog gap returns a same-protocol consent handoff: after explicit permission, the agent can send one structured A2A DataPart with the requested deliverable and only buyer-supplied budget, turnaround, industry, and closest-offer fields. That single entry point evaluates all seventeen published purchase options; the closest-offer field is context only and cannot force a match. The response either finds the offer whose published boundary fits or records a bounded scope-planning signal without creating an order or implementation promise.

Connect

  1. Open the MCP or custom-connector settings in a compatible agent client.
  2. Add a remote Streamable HTTP server using the endpoint above.
  3. Confirm the client discovers buy_matching_service, read-only qualify_task, the find_zinvyl_service and buy_zinvyl_service prompts, and three public resources. An authorized agent should send one sanitized need plus the principal's exact offer allowlist to buy_matching_service. Use buy_service when the exact offer is already known, or the human handoff only when delegation is insufficient. Use get_acceptance_tests when detailed scope review is useful. Use find_service only when the buyer separately authorizes privacy-minimized product-demand telemetry.
  4. Keep tool approval enabled while evaluating the connector. Never place secrets, credentials, regulated data, or raw confidential files in a public routing call.

Recommended tool sequence

01Match read-onlyqualify_task

Send one sanitized outcome and receive the exact offer, price, purchase requirements, and approval handoff—or a catalog gap—without recording buyer demand.

02Review the embedded contractpurchase_requirements

The match already includes exact minor-unit price, currency, terms version, acceptance-test IDs, idempotency rule, and a prefilled human-approval fallback.

03Inspect more detail when neededget_acceptance_tests

Optionally retrieve complete delivery tests, exclusions, and evidence boundaries without creating an intent.

04Match and buy when authorizedbuy_matching_service

Submit one sanitized need, the exact principal-authorized offer allowlist, delegated ceiling, and a stable buyer.request_id. Safe retries return the same bounded intent. Use buy_service when the offer is already selected.

Broader requests and provider access

Authorized buyers and purchasing agents may submit one sanitized broader-work request for private review under the published work-request contract. Human and autonomous-agent providers may submit a review-only capability application through the private matching page. Neither receipt creates a public listing, match, contract, credential, payment, or authorization to begin work. Third-party delivery and payout remain disabled until the required identity, agreement, tax, security, quality, dispute, funding, and payment controls are complete.

Working prompt examples

DISCOVERY

Using Zinvyl, list the available fixed-scope offers. Compare price, delivery window, required inputs, and exclusions. Do not request a payment route.

TRACE READINESS

Using Zinvyl, call qualify_task for this sanitized need: “Check whether one supplied text trace contains the required observable fields for a separately scoped technical review; do not diagnose or repair it.” Return the exact agent-only offer, $0.02-per-pull price, evidence checklist, exclusions, proof link, and Stripe MPP next action without recording payment or funding.

FREIGHT QUALIFICATION

Using Zinvyl, call qualify_task with this sanitized task: “Carrier invoices, bills of lading, and proofs of delivery are manually compared before a load is approved.” Return the recommended offer, price, delivery window, proof link, and next action. Do not diagnose the workflow or record demand telemetry.

UNMET DEMAND

Using Zinvyl, call match_or_capture_demand for this sanitized business need. If an offer matches, return it. Otherwise, record the need only after confirming I am authorized to share it and that it contains no personal data, credentials, payment data, confidential text, or secrets.

AUTHORIZED AGENT PURCHASE

Using Zinvyl, call buy_matching_service once for one sanitized API trace. My principal-authorized allowlist must contain the matched exact offer and my delegated ceiling must cover its published fee, USD currency, and current terms version. Do not claim funding or work until Zinvyl verifies payment.

Inspect proof before purchase

Review the synthetic $0.02 agent trace-readiness verdict, public-data freight audit, synthetic API diagnosis, synthetic venture evidence brief, synthetic interactive prototype, or Zinvyl's owned-brand AI marketing portfolio. Every sample states its evidence boundary and does not imply a paid customer, live demand, or production outcome.

Expected safety behavior

buy_matching_service requires buyer identity, one sanitized task summary, bounded purchase authority with an exact offer allowlist, current terms acceptance, and safe-scope confirmation. buy_service remains the exact-offer path. MCP idempotency uses buyer.request_id in the JSON body, not an HTTP header; an unchanged retry returns the same intent and reservation. A successful purchase call returns selected_payment_route. The Trace Readiness Check is agent-only and requires Stripe MPP; it never returns a human checkout or approval fallback. Other offers may use adaptive delegated-card, hosted-checkout, MPP, or human-approval routes according to their published contract. Only a verified Stripe receipt recorded in the G.B. Industry payment ledger can fund the order. Complete sanitized inputs remain a separate post-funding gate before work starts. See the canonical lifecycle and stop taxonomy for the separate ledgers and exact failure classes.

Reviewer test

No test account or customer data is required. Call list_offers, inspect get_purchase_requirements, and use the compatibility create_purchase_intent tool with example.com contact data and delegated authority set to false. The final call should return authorization_required and funding_state: unverified without creating an intent.

Troubleshooting

The server is not discovered
Confirm the complete HTTPS endpoint is used, the client supports remote Streamable HTTP MCP servers, and outbound HTTPS is permitted.
A tool rejects the request
Use the exact published offer ID. Keep task summaries between 20 and 3,000 characters and remove credentials, tokens, unredacted sensitive or regulated personal data, production secrets, and raw confidential files. Required organization and authorized business-contact fields remain permitted.
No payment route is returned
The agent must have explicit authority for the published offer and fee. When authority is missing, forward the returned human_approval_handoff.url to the human budget owner for an explicit approve-or-decline decision.
Checkout succeeded but status is pending
Do not pay again. Webhook reconciliation can take a short period. Use the returned order-status route; only the signed ledger record determines funding.

Support and policies

Contact Zinvyl. Review the Privacy Policy, Security Statement, Acceptable Use Policy, and Terms before connecting the service to production workflows.

Contact