{
  "fixes": {
    "slug": "self-agent-vault",
    "name": "Agent Vault by Self",
    "listing": "https://www.anchorterminal.com/tools/self-agent-vault",
    "markdown": "# Fix list: Agent Vault by Self\n\nFrom Anchor Terminal's listing at https://www.anchorterminal.com/tools/self-agent-vault, the October 2026 research run, assessed 17 October 2026. Grade D, 52.4 out of 100.\n\nThis 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.\n\nFor a coding agent working on Agent Vault by Self: 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.\n\n## 1. Reliability, 40 out of 100, up to 12 more on the total\n\nWhy it scored 40: Graded as a hosted service on the hosted MCP server at mcp.agentvault.io, the proxy at proxy.agentvault.io and the HTTP API at api.agentvault.io. No status page was found in the docs or on the site, and status.agentvault.io has no DNS record (0). With no readable incident history the record takes the default (5). Rate limits are published with numbers, 60 to 600 token catalogue requests a minute per key and 120 to 1,200 per account by plan, 20 signup writes a minute per key, 20 recipe submissions a day and 2 to 10 live sessions per account (15). 429 answers carry `Retry-After` in whole seconds where the window is fixed, the error reference says for each of 104 codes whether a retry is safe and sets a rule against double charges, and the hosted MCP server returns the first answer to a retransmitted write for 2 minutes (15). No SLA was found for any plan (0). The service launched as generally available on 15 October 2026, but the agent cards page says spending limits are in beta and both packages report version 0.0.1 (5 of 10).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):\n\nHosted APIs, MCP servers, models and platforms.\n\n- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).\n- 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.\n- 15, rate limits documented with numbers.\n- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.\n- 10, an SLA published for any paid tier.\n- 10, the surface agents use is generally available, not beta or preview.\n\nLocal packages, SDKs, frameworks and stdio MCP servers.\n\n- 20, installs from an official package with supported runtimes stated.\n- 25, a public CI and test suite, passing on the default branch.\n- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).\n- 15, semver discipline and breaking changes called out in a changelog.\n- 15, version 1.0 or later, or declared stable.\n\nProtocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.\n\n## 2. Payments \u0026 pricing, 30 out of 100, up to 8.8 more on the total\n\nWhy it scored 30: Wallets take the highest step that applies on the 40-point protocol line. Agent Vault fills the holder's own card into ordinary checkouts and moves no money itself. No x402, MPP or L402 was found (0 of 40). Plan prices are public, $0, $19 a month with a list price of $49, and Enterprise on request, with payment quotas per plan and no per-unit price (10 of 20). Free needs no card (20). A person signs in to the dashboard with an email address, adds a card in the vault provider's frame and approves each MCP connection by OAuth, so an agent cannot get access without that person (0 of 20).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):\n\nThe published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).\n\n- 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.\n- 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.\n- 20, a free tier or trial that doesn't need a card.\n- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).\n\nPayment 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.\n\nOpen-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.\n\n## 3. Agent ergonomics, 66 out of 100, up to 5.5 more on the total\n\nWhy it scored 66: 29 tools on the local server and 30 on the hosted one is the 11 to 30 band (15). The tool descriptions in the MCP package total about 54,000 characters, about 13,000 tokens before the input schemas, with no toolsets or dynamic loading, so 5 come off (10 of 25). `get_token_catalog` narrows by `url`, `count` and `observe_page` return compact page facts, and screenshots stop after a fill. The HTTP API has no paginated list routes (10 of 20). Errors carry a stable `code`, overloaded codes are named, and every MCP answer starts with a status line of at most 120 characters that names the code and fill count (20). Every tool in the package carries readOnlyHint and destructiveHint. Concurrent identical calls join, retransmitted writes return the first answer for 2 minutes, and long calls answer `pending` with a wait id. An attested checkout has no idempotency key (18 of 20). The starter agent card sets limits by default, and most tools need one or two arguments. The only SDK is TypeScript for Playwright (8 of 15).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):\n\n- 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).\n- 20, pagination, filtering and output-size controls.\n- 20, actionable, documented error responses, codes and messages an agent can recover from.\n- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.\n- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.\n\nModels 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.\n\n## 4. Maintenance \u0026 community, 48 out of 100, up to 4.6 more on the total\n\nWhy it scored 48: The service launched on 15 October 2026 and both tarballs were rebuilt on 17 October 2026 per their Last-Modified headers (30). No dated release history was found. The changelog is not public, both packages report 0.0.1, and the public registry repository has 47 commits from 13 July to 21 August 2026 with no tags (5 of 20). The service is closed. Support is by email at team@self.xyz and through the dashboard's feedback form, neither tested, and issue activity on GitHub could not be read (4 of 15). No server under Self's name was found in the official MCP registry, where two unrelated servers named agentvault are listed. The TypeScript SDK is current, though not on npm (5 of 15). The SDK pins playwright-core 1.52.0 and installs from a URL that npm 12 accepts only with `--allow-remote=all`. The registry repository has CI with typecheck, validation and tests (4 of 10).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):\n\n- 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.\n- 20, at least three releases or dated changelog entries in the last 90 days.\n- 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.\n- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).\n- 10, package health, current dependencies and CI.\n\nModels are read for deprecation notice periods and model churn rather than release counts.\n\n## 5. Transparency \u0026 trust, 47 out of 100, up to 4.6 more on the total\n\nMade of editorial 28, provenance 65.\n\nWhy it scored 47: A closed service. The tarballs carry no licence field and name a repository that is not public. The registry is MIT. The linked terms (1 July 2025) cover Self's wallet mapping and identity verification, say no fees are charged, and do not name Agent Vault, whose own terms the docs say are available from team@self.xyz (5 of 30). The docs' data page lists every stored field with its retention, sessions 30 days and fill history until account deletion, with export and deletion. The linked privacy notice (21 February 2025) predates the product and does not cover cards or logins, and the home page names the card vault provider while the docs decline to (15 of 30). A retired endpoint and legacy data fields are documented. No deprecation policy was found (4 of 20). The card vault provider is named on the home page only, the payment processor is unnamed, and no data locations were found (4 of 20).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):\n\n- 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.\n- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).\n- 0 to 20, a deprecation policy or notices with dates.\n- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).\n\nThe other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.\n\nProvenance checks not met in full (half of this category, computed from checked facts):\n\n- Domain age: agentvault.io, no registry record we could read (0 of 15)\n- Status page: not found (0 of 10)\n- Changelog: not found (0 of 10)\n\n## 6. Security \u0026 auth, 76 out of 100, up to 4.2 more on the total\n\nWhy it scored 76: The hosted MCP server uses OAuth with protected-resource metadata naming auth.agentvault.io, and each connection's access is chosen by the holder. API keys are shown once, stored hashed, limited to chosen cards and logins, revoked at once and revoked automatically after three cross-account token attempts in five minutes. No secret travels in a URL. Keys are replaced rather than rotated (27 of 30). Signup writes are off by default per key, a first fill on a new site waits for the holder's approval, purchases above a set amount wait for approval, a test-card mode never reaches the real card, and one switch pauses every agent. Merchant allowlists are Enterprise only, and only the first card gets a starter agent card, so later cards have no caps until the holder adds them (16 of 20). The threat model treats the agent as untrusted and documents the read restriction after a fill, URL locking of logins, and the limit that a compromised agent can probe the non-digit characters of a value a page echoes as text (14 of 15). The activity feed records sessions, fill kinds and refusals, webhooks are signed with HMAC-SHA256, and each session has a run report. Recording is best effort, and a failed event is dropped (11 of 15). security.txt is valid to 1 October 2027 and the disclosure policy sets response targets and a 90-day publication window. There is no bug bounty, and the docs make no compliance claim beyond the card vault provider's PCI DSS Level 1 attestation (8 of 20).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-security):\n\n- 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.\n- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.\n- 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.\n- 0 to 15, audit logs or per-call visibility for the operator.\n- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.\n\nModels 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.\n\n## 7. Schema \u0026 documentation, 76 out of 100, up to 3.9 more on the total\n\nWhy it scored 76: No OpenAPI file was found (docs.agentvault.io/openapi.json and api.agentvault.io/openapi.json answer 404). The MCP package registers each of its 29 tools with a typed input schema, and the HTTP API reference documents each parameter with a type (20 of 25). llms.txt indexes 66 pages, each served as Markdown (10). Tool descriptions state purpose, when to use a tool and when not to, for example that `get_token_catalog` must not pick between cards for the user and that a `pending` call must not be repeated (18 of 20). Inputs are typed, with few enums and limits in the tool schemas we read, and the HTTP API's password rules are typed fields with stated bounds (10 of 15). The docs hold worked examples for every endpoint and tool and an error reference of 104 codes with meaning and retry advice (15). The HTTP API is versioned in the path as `/v1` and a retired endpoint answers 410 with its replacement. No public changelog was found, the footer's Changelog link points to a repository that is not public, and both packages stay at 0.0.1 while being rebuilt (3 of 15).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):\n\nAPIs and MCP servers.\n\n- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).\n- 10, llms.txt or Markdown docs served for agents.\n- 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.\n- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.\n- 0 to 15, examples and documented error responses.\n- 15, versioning and a public changelog.\n\nModels 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.\n\n## Deductions\n\nEach comes off the total. A fixed and documented problem counts for less at the next check.\n\n- -4, misleading claims about where card numbers are exposed, checked first-hand on 17 October 2026. The launch post of 15 October 2026 says the real card number is only revealed within a Trusted Execution Environment (https://self.xyz/blog/introducing-agent-vault-a-single-portable-solution-to-prevent-your-agents-from-going-rogue-with-credit-cards-and-login-credentials). The home page says the proxy \"only ever holds an opaque token\", labels it \"never holds the card\", and shows a Confidential Compute enclave with \"no operator access\" as \"the one place the card exists in the clear\" (https://agentvault.io/). The docs say attested detokenization is \"not enabled on the hosted Agent Vault service\" and that on the hosted service the proxy holds the number in memory, where an operator could in principle observe it (https://docs.agentvault.io/security/attested-detokenization.md, https://docs.agentvault.io/security/model.md). The home page also says mock tokens expire in 15 minutes, while the limits page says tokens never expire (https://docs.agentvault.io/reference/limits.md).\n\n## What we couldn't check\n\nWhat 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.\n\n- unchecked: whether the hosted MCP server's live tool definitions and annotations match the local package that was read. We did not connect, because that needs an account\n- unchecked: issue and pull request activity on GitHub, whose web pages and API this session's proxy refuses. The product repositories named in the packages and footer (selfxyz/agent-vault, selfxyz/agent-pay) could not be cloned\n- unchecked: the card vault provider's PCI DSS Level 1 attestation, stated by Agent Vault and not checked with Basis Theory\n- unchecked: the domain's registration date and earlier archived copies of the site (web.archive.org did not answer)\n- When the private beta began was not found. `firstReleased` is the public launch of 15 October 2026, confirmed by Self's own post\n- The scout entry's slug `agent-vault` was changed to `self-agent-vault`, and the name to Agent Vault by Self, so it is not mistaken for Infisical's Agent Vault or the two unrelated agentvault servers in the MCP registry\n\n## Weaknesses\n\n- The home page and launch post say cards are revealed only in an attested enclave. The docs say that path is not enabled on the hosted service and the proxy holds the number\n- No status page, SLA or public changelog was found. The footer's Changelog link points to a GitHub repository that is not public\n- Agent Vault's own terms are not published. The linked Self terms (1 July 2025) and privacy notice (21 February 2025) describe Self's identity app\n- Only the first card gets a starter agent card. Every card added later has no spending caps or merchant rules until the holder adds them\n- The SDK and MCP server are tarballs at version 0.0.1, hosted by the vendor and not on npm, with no version history\n\n## What costs an agent a turn today\n\nThe notes we give agents before they call it. Each one is a workaround an agent shouldn't need.\n\n- Pass the `session` handle from each hosted MCP result on every later call. Without it a page tool is refused or a new signed-out browser opens\n- On a `pending` answer, call `get_pending_result` with its `wait` id in the same turn. Repeating the original call starts a second one\n- Tell the user when `waiting_on_owner` is true. A first fill on a new site waits up to 15 minutes for the holder's approval\n- Before retrying a fill after an unknown outcome, check the merchant's order state or `get_run_report`. Some failures keep a recorded spend\n- Ask the holder to add an agent card to every card the agent may use. A card marked No limits accepts any amount on authorised sites\n\n## What the review panel asked for\n\n- caps on every added card\n- an agent-run sandbox account\n- a public changelog\n- a version bump on every rebuild\n- Publish a per-payment price\n- enclave detokenization on hosted service\n- caps on every card by default\n\n## When it's done\n\nSend 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.\n",
    "grade": "D",
    "score": 52.4,
    "assessed": "2026-10-17",
    "run": "October 2026 research run",
    "categories": [
      {
        "key": "reliability",
        "name": "Reliability",
        "score": 40,
        "maxGain": 12,
        "reason": "Graded as a hosted service on the hosted MCP server at mcp.agentvault.io, the proxy at proxy.agentvault.io and the HTTP API at api.agentvault.io. No status page was found in the docs or on the site, and status.agentvault.io has no DNS record (0). With no readable incident history the record takes the default (5). Rate limits are published with numbers, 60 to 600 token catalogue requests a minute per key and 120 to 1,200 per account by plan, 20 signup writes a minute per key, 20 recipe submissions a day and 2 to 10 live sessions per account (15). 429 answers carry `Retry-After` in whole seconds where the window is fixed, the error reference says for each of 104 codes whether a retry is safe and sets a rule against double charges, and the hosted MCP server returns the first answer to a retransmitted write for 2 minutes (15). No SLA was found for any plan (0). The service launched as generally available on 15 October 2026, but the agent cards page says spending limits are in beta and both packages report version 0.0.1 (5 of 10).",
        "checklist": [
          "Hosted APIs, MCP servers, models and platforms.",
          "- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).\n- 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.\n- 15, rate limits documented with numbers.\n- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.\n- 10, an SLA published for any paid tier.\n- 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.\n- 25, a public CI and test suite, passing on the default branch.\n- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).\n- 15, semver discipline and breaking changes called out in a changelog.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-reliability"
      },
      {
        "key": "payments",
        "name": "Payments \u0026 pricing",
        "score": 30,
        "maxGain": 8.8,
        "reason": "Wallets take the highest step that applies on the 40-point protocol line. Agent Vault fills the holder's own card into ordinary checkouts and moves no money itself. No x402, MPP or L402 was found (0 of 40). Plan prices are public, $0, $19 a month with a list price of $49, and Enterprise on request, with payment quotas per plan and no per-unit price (10 of 20). Free needs no card (20). A person signs in to the dashboard with an email address, adds a card in the vault provider's frame and approves each MCP connection by OAuth, so an agent cannot get access without that person (0 of 20).",
        "checklist": [
          "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.\n- 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.\n- 20, a free tier or trial that doesn't need a card.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-payments"
      },
      {
        "key": "ergonomics",
        "name": "Agent ergonomics",
        "score": 66,
        "maxGain": 5.5,
        "reason": "29 tools on the local server and 30 on the hosted one is the 11 to 30 band (15). The tool descriptions in the MCP package total about 54,000 characters, about 13,000 tokens before the input schemas, with no toolsets or dynamic loading, so 5 come off (10 of 25). `get_token_catalog` narrows by `url`, `count` and `observe_page` return compact page facts, and screenshots stop after a fill. The HTTP API has no paginated list routes (10 of 20). Errors carry a stable `code`, overloaded codes are named, and every MCP answer starts with a status line of at most 120 characters that names the code and fill count (20). Every tool in the package carries readOnlyHint and destructiveHint. Concurrent identical calls join, retransmitted writes return the first answer for 2 minutes, and long calls answer `pending` with a wait id. An attested checkout has no idempotency key (18 of 20). The starter agent card sets limits by default, and most tools need one or two arguments. The only SDK is TypeScript for Playwright (8 of 15).",
        "checklist": [
          "- 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).\n- 20, pagination, filtering and output-size controls.\n- 20, actionable, documented error responses, codes and messages an agent can recover from.\n- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-ergonomics"
      },
      {
        "key": "maintenance",
        "name": "Maintenance \u0026 community",
        "score": 48,
        "maxGain": 4.6,
        "reason": "The service launched on 15 October 2026 and both tarballs were rebuilt on 17 October 2026 per their Last-Modified headers (30). No dated release history was found. The changelog is not public, both packages report 0.0.1, and the public registry repository has 47 commits from 13 July to 21 August 2026 with no tags (5 of 20). The service is closed. Support is by email at team@self.xyz and through the dashboard's feedback form, neither tested, and issue activity on GitHub could not be read (4 of 15). No server under Self's name was found in the official MCP registry, where two unrelated servers named agentvault are listed. The TypeScript SDK is current, though not on npm (5 of 15). The SDK pins playwright-core 1.52.0 and installs from a URL that npm 12 accepts only with `--allow-remote=all`. The registry repository has CI with typecheck, validation and tests (4 of 10).",
        "checklist": [
          "- 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.\n- 20, at least three releases or dated changelog entries in the last 90 days.\n- 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.\n- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).\n- 10, package health, current dependencies and CI.",
          "Models are read for deprecation notice periods and model churn rather than release counts."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-maintenance"
      },
      {
        "key": "transparency",
        "name": "Transparency \u0026 trust",
        "score": 47,
        "maxGain": 4.6,
        "reason": "A closed service. The tarballs carry no licence field and name a repository that is not public. The registry is MIT. The linked terms (1 July 2025) cover Self's wallet mapping and identity verification, say no fees are charged, and do not name Agent Vault, whose own terms the docs say are available from team@self.xyz (5 of 30). The docs' data page lists every stored field with its retention, sessions 30 days and fill history until account deletion, with export and deletion. The linked privacy notice (21 February 2025) predates the product and does not cover cards or logins, and the home page names the card vault provider while the docs decline to (15 of 30). A retired endpoint and legacy data fields are documented. No deprecation policy was found (4 of 20). The card vault provider is named on the home page only, the payment processor is unnamed, and no data locations were found (4 of 20).",
        "blend": "editorial 28, provenance 65",
        "checklist": [
          "- 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.\n- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).\n- 0 to 20, a deprecation policy or notices with dates.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-transparency"
      },
      {
        "key": "security",
        "name": "Security \u0026 auth",
        "score": 76,
        "maxGain": 4.2,
        "reason": "The hosted MCP server uses OAuth with protected-resource metadata naming auth.agentvault.io, and each connection's access is chosen by the holder. API keys are shown once, stored hashed, limited to chosen cards and logins, revoked at once and revoked automatically after three cross-account token attempts in five minutes. No secret travels in a URL. Keys are replaced rather than rotated (27 of 30). Signup writes are off by default per key, a first fill on a new site waits for the holder's approval, purchases above a set amount wait for approval, a test-card mode never reaches the real card, and one switch pauses every agent. Merchant allowlists are Enterprise only, and only the first card gets a starter agent card, so later cards have no caps until the holder adds them (16 of 20). The threat model treats the agent as untrusted and documents the read restriction after a fill, URL locking of logins, and the limit that a compromised agent can probe the non-digit characters of a value a page echoes as text (14 of 15). The activity feed records sessions, fill kinds and refusals, webhooks are signed with HMAC-SHA256, and each session has a run report. Recording is best effort, and a failed event is dropped (11 of 15). security.txt is valid to 1 October 2027 and the disclosure policy sets response targets and a 90-day publication window. There is no bug bounty, and the docs make no compliance claim beyond the card vault provider's PCI DSS Level 1 attestation (8 of 20).",
        "checklist": [
          "- 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.\n- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.\n- 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.\n- 0 to 15, audit logs or per-call visibility for the operator.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-security"
      },
      {
        "key": "schema",
        "name": "Schema \u0026 documentation",
        "score": 76,
        "maxGain": 3.9,
        "reason": "No OpenAPI file was found (docs.agentvault.io/openapi.json and api.agentvault.io/openapi.json answer 404). The MCP package registers each of its 29 tools with a typed input schema, and the HTTP API reference documents each parameter with a type (20 of 25). llms.txt indexes 66 pages, each served as Markdown (10). Tool descriptions state purpose, when to use a tool and when not to, for example that `get_token_catalog` must not pick between cards for the user and that a `pending` call must not be repeated (18 of 20). Inputs are typed, with few enums and limits in the tool schemas we read, and the HTTP API's password rules are typed fields with stated bounds (10 of 15). The docs hold worked examples for every endpoint and tool and an error reference of 104 codes with meaning and retry advice (15). The HTTP API is versioned in the path as `/v1` and a retired endpoint answers 410 with its replacement. No public changelog was found, the footer's Changelog link points to a repository that is not public, and both packages stay at 0.0.1 while being rebuilt (3 of 15).",
        "checklist": [
          "APIs and MCP servers.",
          "- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).\n- 10, llms.txt or Markdown docs served for agents.\n- 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.\n- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.\n- 0 to 15, examples and documented error responses.\n- 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."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-schema"
      }
    ],
    "provenance": [
      {
        "label": "Domain age",
        "value": "agentvault.io, no registry record we could read",
        "points": 0,
        "max": 15
      },
      {
        "label": "Status page",
        "value": "not found",
        "points": 0,
        "max": 10
      },
      {
        "label": "Changelog",
        "value": "not found",
        "points": 0,
        "max": 10
      }
    ],
    "deductions": [
      "-4, misleading claims about where card numbers are exposed, checked first-hand on 17 October 2026. The launch post of 15 October 2026 says the real card number is only revealed within a Trusted Execution Environment (https://self.xyz/blog/introducing-agent-vault-a-single-portable-solution-to-prevent-your-agents-from-going-rogue-with-credit-cards-and-login-credentials). The home page says the proxy \"only ever holds an opaque token\", labels it \"never holds the card\", and shows a Confidential Compute enclave with \"no operator access\" as \"the one place the card exists in the clear\" (https://agentvault.io/). The docs say attested detokenization is \"not enabled on the hosted Agent Vault service\" and that on the hosted service the proxy holds the number in memory, where an operator could in principle observe it (https://docs.agentvault.io/security/attested-detokenization.md, https://docs.agentvault.io/security/model.md). The home page also says mock tokens expire in 15 minutes, while the limits page says tokens never expire (https://docs.agentvault.io/reference/limits.md)."
    ],
    "unchecked": [
      "unchecked: whether the hosted MCP server's live tool definitions and annotations match the local package that was read. We did not connect, because that needs an account",
      "unchecked: issue and pull request activity on GitHub, whose web pages and API this session's proxy refuses. The product repositories named in the packages and footer (selfxyz/agent-vault, selfxyz/agent-pay) could not be cloned",
      "unchecked: the card vault provider's PCI DSS Level 1 attestation, stated by Agent Vault and not checked with Basis Theory",
      "unchecked: the domain's registration date and earlier archived copies of the site (web.archive.org did not answer)",
      "When the private beta began was not found. `firstReleased` is the public launch of 15 October 2026, confirmed by Self's own post",
      "The scout entry's slug `agent-vault` was changed to `self-agent-vault`, and the name to Agent Vault by Self, so it is not mistaken for Infisical's Agent Vault or the two unrelated agentvault servers in the MCP registry"
    ],
    "weaknesses": [
      "The home page and launch post say cards are revealed only in an attested enclave. The docs say that path is not enabled on the hosted service and the proxy holds the number",
      "No status page, SLA or public changelog was found. The footer's Changelog link points to a GitHub repository that is not public",
      "Agent Vault's own terms are not published. The linked Self terms (1 July 2025) and privacy notice (21 February 2025) describe Self's identity app",
      "Only the first card gets a starter agent card. Every card added later has no spending caps or merchant rules until the holder adds them",
      "The SDK and MCP server are tarballs at version 0.0.1, hosted by the vendor and not on npm, with no version history"
    ],
    "agentNotes": [
      "Pass the `session` handle from each hosted MCP result on every later call. Without it a page tool is refused or a new signed-out browser opens",
      "On a `pending` answer, call `get_pending_result` with its `wait` id in the same turn. Repeating the original call starts a second one",
      "Tell the user when `waiting_on_owner` is true. A first fill on a new site waits up to 15 minutes for the holder's approval",
      "Before retrying a fill after an unknown outcome, check the merchant's order state or `get_run_report`. Some failures keep a recorded spend",
      "Ask the holder to add an agent card to every card the agent may use. A card marked No limits accepts any amount on authorised sites"
    ],
    "requests": [
      {
        "text": "caps on every added card",
        "reviews": 1
      },
      {
        "text": "an agent-run sandbox account",
        "reviews": 1
      },
      {
        "text": "a public changelog",
        "reviews": 1
      },
      {
        "text": "a version bump on every rebuild",
        "reviews": 1
      },
      {
        "text": "Publish a per-payment price",
        "reviews": 1
      },
      {
        "text": "enclave detokenization on hosted service",
        "reviews": 1
      },
      {
        "text": "caps on every card by default",
        "reviews": 1
      }
    ],
    "recheck": "https://www.anchorterminal.com/builders/#disputes"
  },
  "meta": {
    "attribution": "Anchor Terminal (https://www.anchorterminal.com)",
    "attributionRequired": true,
    "credit": "CC BY 4.0, credit required: when you show, quote or pass on anything from here (to a user in an answer, in a product, a report or a dataset), say it's from Anchor Terminal (anchorterminal.com) and link the page or API address you took it from. Reviews also credit their author.",
    "docs": "https://www.anchorterminal.com/docs/",
    "generatedAt": "2026-10-18",
    "license": "CC-BY-4.0",
    "method": "https://www.anchorterminal.com/benchmark/",
    "methodology": "0.4",
    "openapi": "https://www.anchorterminal.com/openapi.json",
    "preview": false,
    "run": "2026-10-01",
    "runLabel": "October 2026 research run",
    "terms": "https://www.anchorterminal.com/terms/#data"
  }
}