SOURCE COVERAGE & PRACTICAL LIMITS
Full product guide
Source contract captured from build 1.0.1. Presentation refreshed 2026-11-21. Example outputs retain their original sample context; they are not live results.
A2A Agent Card Monitor – Skills & Contract Changes
Did a remote AI agent change its public skills, interfaces or authentication contract?
Spot public AI agent contract changes before you update your integration. Keep skill additions, removed skills, interface changes and advertised authentication changes in source-linked JSON or CSV.
Start with a small useful result
For AI integration agencies, agent directory operators and developers maintaining a shortlist of public A2A services. No API key, account login or proxy is required. Press Start to collect at most two real public observations. Leave comparison off for the first exploration. Every row includes a stable recordId, title, url, sourceUrl and observedAt.
What you can do with the results
Each row represents one validated public card and includes skills, interfaces, capabilities, input/output modes, provider metadata and selected security scheme declarations, including API key location, advertised OAuth scopes, flow endpoint URLs and PKCE flags. skillCount counts advertised skills, not successful calls. Modern supportedInterfaces and legacy url/preferredTransport fields are normalized, while cardLayout keeps their origin visible and protocolFamily derives from the declared versions. Interface preference order and tenant routing values are preserved. Extra custom fields, call examples, signatures and secret-like scheme values are not copied.
Compare two public Agent Card contracts
Read ACN Preflight and AgentCard.net Toolkit public metadata. This does not call any advertised skill.
{
"cardUrls": [
"https://preflight.abacore.net/.well-known/agent-card.json",
"https://www.agentcard.net/.well-known/agent-card.json"
],
"maxResults": 2
}
Inspect a domain’s public discovery card
Resolve AgentCard.net to its standard well-known card and export its declared skills and interfaces.
{
"cardUrls": [
"https://www.agentcard.net"
],
"maxResults": 1
}
Start a public agent contract change watch
Establish a private baseline for the two public example cards; later complete runs identify skill and contract changes.
{
"cardUrls": [
"https://preflight.abacore.net/.well-known/agent-card.json",
"https://www.agentcard.net/.well-known/agent-card.json"
],
"maxResults": 2,
"comparePrevious": true,
"snapshotName": "agent-contract-watch"
}
Changes you can verify
On complete subsequent observations, addedSkills, removedSkills and changedSkills describe card-level changes. interfacesChanged, authChanged and capabilitiesChanged identify contract differences. Removed skills mean removed from the newly retrieved complete card; an unavailable card never establishes that an agent or skill was removed. Authentication requirements absent from a card remain null; an explicit empty list remains distinct.
Set comparePrevious: true and a stable snapshotName to retain a private baseline for the selected scope. The first complete run marks observations new; later runs mark them changed or unchanged with changedFields and the previous normalized observation. The baseline belongs to the account running the Actor. A different source list or comparison scope creates separate history.
onlyChanges: true enables comparison and suppresses unchanged rows; a successful repeat can legitimately save zero rows. Source failures, truncated output, a deadline or a spending limit preserve the prior baseline. A concurrent-history change detected before commit is reported rather than overwritten; avoid overlapping runs for the same watchlist because the storage service does not provide an atomic compare-and-swap transaction.
Coverage and limits
- The default reads ACN Preflight and AgentCard.net Toolkit, two independent public examples. They are third-party services, not Mayd It services. Their metadata can change.
- Supply public domains or the exact
/.well-known/agent-card.json path. Legacy /.well-known/agent.json is supported only when explicitly selected; no hidden fallback requests.
- Each source uses a read-only GET. Standard HTTP(S) ports only; URL credentials and query strings are refused. All resolved IP addresses must be public, and validated addresses are pinned for the connection to prevent DNS rebinding.
- Response bodies are limited to 512 KiB before and after decompression. Redirects are not followed; correct the public source URL when a card moves.
- A declared private interface is recorded with a warning and never contacted. A signature count is presence metadata only:
signatureVerification is always not_performed.
- This is a bounded contract monitor, not a complete A2A conformance validator, agent marketplace, trust score, security audit, or uptime guarantee.
- At most 20 targets and 20 saved observations per run. Requests use a 120-second maximum run budget and the platform’s earlier deadline.
- All-source failure fails honestly. Source errors and access problems are written to
SUMMARY, never fabricated as paid dataset rows. A partial run retains useful rows and clearly reports incomplete coverage.
Output, pricing and scheduling
Export the dataset as JSON, CSV or Excel. Read SUMMARY for source coverage, limits and history status; open RUN_REPORT for a readable overview. For nested checks or skills, JSON preserves the full structure.
Base pricing is $0.002 per confirmed saved observation, plus $0.00005 per Actor start at the default 256 MB. Standard plan-tier discounts may apply to results. A row containing several checks or skills is still one observation. Failed sources and unchanged results suppressed by onlyChanges do not incur a result charge. Apify platform usage may also apply under your plan. The Actor checks remaining result allowance before saving, and the last accepted row at the limit is counted correctly.
For recurring checks, save a task with your watchlist and use Apify Schedules at a cadence appropriate to the upstream source. Configure integrations yourself if you need to deliver results elsewhere; this Actor sends no messages and creates no schedules.
Source and interpretation
The A2A specification defines public Agent Card discovery and metadata. ACN Preflight card and AgentCard.net card are the documented examples. This is an independent tool, not an official A2A product.
Use the Actor’s Issues tab to report source changes or request a supported input example.
Companion workflows
Each Actor runs separately with its own inputs, authorization and pricing. These instructions do not configure automatic execution.
Maintain separate evidence for A2A agent contracts and MCP server registrations.
- Review changed skills, versions and interface declarations in the public Agent Card.
- Run the MCP registry monitor for separately selected relevant MCP servers.
- Use both exports to maintain an integration inventory with source and observation timestamps.
A2A and MCP are different protocols. A declared capability or registry entry is not proof that execution works; no agent invocation is performed.
Review API contract changes for services that also expose public OpenAPI descriptions.
- Identify a changed agent's public service interface from your own integration documentation.
- Provide a verified public OpenAPI document URL to the OpenAPI monitor.
- Compare the two contract-change reports before testing your integration.
An Agent Card URL is not an OpenAPI document. Do not automatically follow untrusted card links or transmit credentials.