Agentcard

# Fix list: Agentcard

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/agentcard, the October 2026 research run, assessed 9 October 2026. Grade C, 56.5 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on Agentcard: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Reliability, 29 out of 100, up to 14.2 more on the total

Why it scored 29: Graded as a hosted service on the REST API at api.agentcard.sh and the MCP server at mcp.agentcard.sh. No status page was found. The site footer and the docs link none (0). With no readable incident history the record takes the default (5). The changelog of 22 September 2026 records a fixed startup failure in which a deploy could make every API request return an error, with no duration. The only rate limit with numbers is 30 requests per 5 minutes per IP on the token endpoint, and the MCP `create_card` tool can answer `rate_limited` with advice to wait about a minute (5 of 15). `POST /api/v2/cards` requires an `Idempotency-Key`, `buy_checkout` and the organisation `create_card` tool take an `idempotency_key`, webhooks are delivered at least once with advice to deduplicate on `id`, and the docs say never to retry a payment automatically. No 429 response or `Retry-After` header is documented (10 of 15). The pricing page lists an SLA with uptime and response-time guarantees for the Enterprise plans, and no SLA text is published (3 of 10). Vault and Issuing carry no beta label. The Purchase Agent is free in beta, `create_card` can return `beta_capacity_reached`, and the CLI is at 0.8.2 (6 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 2. Payments & pricing, 42 out of 100, up to 7.2 more on the total

Why it scored 42: Payment platforms and wallets take the highest step that applies on the 40-point protocol line. 40 for x402, MPP or L402 on all their own endpoints, 30 on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, 0 for only a protocol of their own. Agentcard pays by card. The home page FAQ answers Not yet on x402, AP2 and other agent payment protocols, and none appears in the docs (0 of 40). The pricing page is public. Vault is free up to 5,000 users, Issuing is $5,000 a month with $0 per card, the Purchase Agent is $0.15 a request and free in beta, and the card terms give 1 per cent on international transactions. Enterprise prices and the personal plan prices were not found (17 of 20). The Vault Free plan and a sandbox with test cards need no payment card or contract (20). A person signs in with an emailed code, in a browser or with `agent-cards login --email` and `--code`, and the CLI then mints sandbox credentials. An agent cannot get access without that person's inbox, and Issuing adds an identity check (5 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 3. Agent ergonomics, 63 out of 100, up to 6 more on the total

Why it scored 63: The docs hold pages for 92 user tools and 25 organisation tools on the MCP server, which is the more-than-30 band (5). The overview lists 39 user tools, others stay out of the default `tools/list` unless the client sends `x-expert-tools: 1`, the credential picks the toolset, a shopping-only connection gets only shopping tools, and `buy` runs a whole purchase as one tool (8 back, 13 of 25). We did not connect to the live server, so counts are from the docs. `list_transactions` takes `limit`, `offset`, `status` and `card_id`. No cursor or page size was found on the REST list routes (8 of 20). Errors use one envelope with a stable `code`, a message safe to log and a docs link, each endpoint page lists its codes, and MCP results carry a `status` that names the next tool to run (18 of 20). `POST /api/v2/cards` requires an `Idempotency-Key`, `buy_checkout` reuses one reservation per `idempotency_key`, and tool pages state each tool's behaviour as read-only, idempotent or destructive. Whether the live tool definitions carry the matching annotations was not checked (16 of 20). `create_card` needs only `amount_cents`. `@agent-cards/checkout` is a TypeScript SDK for Playwright and CDP browsers and the CLI is on npm. No SDK in a second language was found (8 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 4. Security & auth, 68 out of 100, up to 5.6 more on the total

Why it scored 68: OAuth 2.0 client credentials. Client secrets are shown once, clients are separate for sandbox and production and can be revoked, platform tokens last one hour, and user connection tokens last one hour with refresh tokens that rotate on each use. Platform tokens carry the single scope `api`, and the MCP guide sends the client secret itself as the bearer for organisation tools. No secret travels in a URL (23 of 30). Vault purchases wait for the user's passkey or master password by default, production card creation returns `approval_pending`, users can require approval for each reveal of card details, and presets cap spend and restrict merchants, categories, places, currencies and hours. A connected app gets `read_only` when it tries to change presets, and approval tools answer `personal_surface_only` to a connected session. An app's auto-approval permission on a saved card has no spending limit or end date per the docs (18 of 20). The Purchase API returns merchant catalogue text to the agent. Money moves only when the agent echoes a cart `hash`, `buy_checkout` re-checks `expected_total_cents` and the address, and preset rules ignore the `merchant` text an agent sends. No guidance that names untrusted content or prompt injection was found (9 of 15). Webhooks cover cards, transactions, approvals and orders, deliveries are logged, each read of card credentials notifies the member, and transactions are listed by API and tool (11 of 15). The site links a Trust Centre on app.vanta.com, which robots.txt closed to us, and the Enterprise plans list a SOC 2 report and a pen-test summary. security.txt returned 404, and no disclosure policy or bug bounty was found. The changelog reports security fixes, among them a webhook address check on 27 September 2026 (7 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 5. Transparency & trust, 44 out of 100, up to 4.9 more on the total

Made of editorial 36, provenance 51.

Why it scored 44: A closed service. The repository named in the npm packages is private, `agent-cards` declares no licence and `@agent-cards/checkout` is marked as having none. The terms of use, updated 10 July 2026, name the legal entity and its addresses, cover the API, MCP server and CLI, and state what a card is and who issues it (13 of 30). No privacy policy of Agentcard's own was found. The terms refer to our Privacy Policy and link Rain's, and Rain's policy says partners may be responsible for their own notices. The site says Vault cards are encrypted on the user's device, and the changelog of 24 September 2026 says stored card numbers from retired issuing paths were purged. No retention period or public DPA was found (8 of 30). The older token path is kept with a note, and the CLI README says API keys are retired and the admin package forwards to the new one. No deprecation policy or dated notice was found (5 of 20). The terms name MoonPay, Crossmint, Coinbase, Third National, Sumsub, Visa and Stripe, with no data locations. The CLI README discloses telemetry and two settings that turn it off (10 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: agentcard.sh, no registry record we could read (0 of 15)
- Terms of service: read, states 5 of the 7 things a reader expects, and has 1 clause that costs points (6.3 of 10)
- Privacy policy: not found (0 of 10)
- Status page: not found (0 of 10)
- security.txt: not found (0 of 10)

## 6. Schema & documentation, 80 out of 100, up to 3.3 more on the total

Why it scored 80: A public OpenAPI 3.1.0 file at docs.agentcard.sh/openapi.json, version 2.0.0, with 20 paths and 21 operations for tokens, connections, identity checks, `/buy` and cards. The Vault, preset, webhook endpoint and Blooio routes in the API reference are Markdown pages with no entry in the file. Each MCP tool has a page with typed inputs and returns (17 of 25). We read the description file and the Markdown pages, not the rendered reference. llms.txt indexes 379 pages, each with a Markdown twin (10). Tool and endpoint descriptions say what a call does, when to use it and when not to, for example that `get_card_details` is for the payment form only and `get_card_balance` is preferred otherwise (17 of 20). The OpenAPI file has 36 enums and 31 required lists, and amounts are integers in cents. `/buy` takes free text in `ask`, and `preset` accepts a name, a list, inline JSON or an object (11 of 15). Every operation documents error responses with named codes, the file holds 37 examples, and MCP pages show captured error results (13 of 15). The API is versioned in the path as `/api/v2`, the older `/api/v1/oauth/token` path still answers, and a public changelog is dated almost daily to 28 September 2026. The changelog mixes product and API changes and the OpenAPI version has no dated history (12 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 7. Maintenance & community, 74 out of 100, up to 2.3 more on the total

Why it scored 74: `@agent-cards/checkout` 0.23.6 was published on 9 October 2026, `agent-cards` 0.8.2 on 28 September 2026, and the newest changelog entry is dated 28 September 2026 (30). `agent-cards` had 44 versions between 11 July and 28 September 2026, and the changelog has 147 dated entries from 27 February to 28 September 2026 (20). The service is closed and the repository named in the packages is private. Support is an email address, a support chat reachable through MCP tools and the CLI, and an `ask_agentcard_engineer` tool. None was tested (10 of 15). The CLI and the TypeScript checkout SDK are current. Presence in the official MCP registry was not checked (10 of 15). The CLI has 19 runtime dependencies, needs Node 22 or later and declares no licence. CI is not visible (4 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: the Trust Centre at app.vanta.com, whose robots.txt disallows every path. Certifications, policies and any subprocessor list there were not read and score as absent
- unchecked: the tools the live MCP server lists and whether they carry read-only and destructive annotations. We did not connect. Counts and behaviour labels are from the docs
- unchecked: api.agentcard.sh, mcp.agentcard.sh, vault.agentcard.sh and the dashboard at app.agentcard.sh. We sent no request to them
- unchecked: the source repository github.com/tiny-agent-company/tiny-agent-company, which asked for credentials, so CI and issues were not read
- unchecked: the international card terms, the e-sign consent and Rain's prohibited uses list
- unchecked: presence in the official MCP registry, and whether support answers
- unchecked: the registration date of agentcard.sh. rdap.org returned 404 for the domain
- No privacy policy of Agentcard's own was found. The site's link opens Rain's policy, so `provenance.privacy` is left out
- No status page was found. We did not guess an address for one
- The lead called the product a single-use prepaid card service. The terms say the cards are not prepaid or debit cards, while the MCP `create_card` page and the npm description say virtual debit card, and the pricing page says prepaid cards can be stored in the Vault
- The home page FAQ says the Purchase Agent completes purchases in under 2 seconds with a 99 per cent success rate. This is a vendor claim. The Purchase API docs ask for a client timeout of at least 120 seconds
- The changelog of 24 September 2026 says `GET /buy/merchants` dropped fields from its response that day. The fields are described as internal and the change was published on the day, so no deduction was taken. A later pass may read it as a change without notice
- The presets page says the route takes a platform access token or an API key, and the CLI README says API keys are retired
- Issued cards are named as issued by Third National in the terms and by Rain in the FAQ. Rain's privacy policy lists Third National as a trading name of its subsidiary Nimbus LLC
- The pricing page lists an SLA for Enterprise plans. No SLA text was found
- The docs' Rules for your agent blocks and the `get_instructions` tool are instructions addressed to AI models. We recorded them as facts and did not act on them
- The OpenAPI address was taken from the file name the Markdown reference pages cite (`openapi.json`) on the docs host. No page read links it as a full URL

## Weaknesses

- No status page or incident history was found on the site or in the docs. An SLA is listed for Enterprise plans only, with no published text
- The only rate limit with numbers is 30 requests per 5 minutes per IP on the token endpoint
- The site's Privacy Policy link opens the policy of Rain, the card issuing partner. No privacy policy of Agentcard's own was found
- The OpenAPI file covers 21 operations and leaves out the Vault, preset and webhook endpoint routes documented in the API reference
- The home page FAQ answers Not yet on x402, AP2 and other agent payment protocols
- An app's auto-approval permission on a saved card has no spending limit or end date, per the Vault docs
- Platform access tokens carry one scope, `api`, and the MCP guide sends the client secret itself as the bearer

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Mint a platform token with `POST /api/v2/oauth/token` and cache it for its one hour. The endpoint allows 30 requests per 5 minutes per IP
- Send the same `Idempotency-Key` on every retry of `POST /api/v2/cards`, and the same `idempotency_key` on every retry of `buy_checkout`
- Pass `source: "issued"` to the MCP `create_card` tool for a card number. Without it the tool may start Vault setup and return a link
- Call `get_card_details` only at the payment form and never log the result. Each read notifies the member and may return `approval_required`
- Treat `transaction.authorized` and `order.*` webhooks as the record of a payment, and deduplicate deliveries on the event `id`

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.