withHuman

# Fix list: withHuman

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/withhuman, the October 2026 research run, assessed 8 October 2026. Grade D, 51.8 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 withHuman: 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, 16 out of 100, up to 16.8 more on the total

Why it scored 16: Graded as a hosted service. No status page is linked from the site, the docs or the support page, and status.withhuman.ai did not connect (0), so there is no incident history to read (5). No request rate limits are published. 429 is documented only for the plan's limit on pending requests or active agents, with no numbers in the docs (0). `Idempotency-Key` is required on four write commands with documented replay and conflict behaviour, and the long-poll guide says to call again after a dropped connection, but no Retry-After or backoff guidance was found for 429 (11 of 15). The terms give no uptime guarantee outside an enterprise agreement (0). The terms say the service is in early access and the API document is version 0.1.0 (0).

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, 30 out of 100, up to 8.8 more on the total

Why it scored 30: No machine payment protocol (0). Public plan prices, Free at $0 for up to 5 members and Scale at $20 per user a month, with no per-request price and unpublished monthly allowances for gateway calls and phone messages (10). The Free plan and trial need no subscription per the billing docs. We did not sign up to confirm that no card is asked for (20). A person signs up in a browser and approves each agent's enrolment. Provisioner tokens then issue instance credentials by API (0).

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. Security & auth, 73 out of 100, up to 4.7 more on the total

Why it scored 73: Purpose-limited credentials. An agent credential can only create and read its own requests, a provisioner token only manages instances, personal keys can be narrowed and expire after 90 days by default, and the MCP server uses OAuth with PKCE, dynamic client registration and 35 scopes. Choosing specific actions is limited to the hosted paid plans (28 of 30). Agents and organisation keys can never decide a request, sensitive grants need a recent sign-in, and the docs say the required-MFA settings don't yet work with the built-in sign-in methods (16 of 20). Reviewers see agent reasoning as a claim separate from the arguments, `server` is documented as an unverified label, and notifications never carry arguments (11 of 15). Append-only audit log with search by API, and every key's changes recorded under the key, in the hosted edition (13 of 15). The footer says SOC 2 Type II with no report, auditor or trust page. No security.txt (404 on both hosts), disclosure policy or bounty found (5 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.

## 4. Maintenance & community, 49 out of 100, up to 4.5 more on the total

Why it scored 49: The CLI's `latest` build files were last modified on 8 October 2026, and the sitemap dates docs changes to 6 October (30). No versioned release notes or changelog were found. The sitemap shows docs edits on eight dates between 22 September and 6 October, which we count in part (5 of 20). Closed service with email support said to reply within one working day and a Discord for the protocol. The linked GitHub repositories returned 404 (6 of 15). No official SDK found. The CLI is the only client (5 of 15). CLI builds for four platforms ship with SHA256SUMS, and no public CI was found (3 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.

## 5. Agent ergonomics, 74 out of 100, up to 4.2 more on the total

Why it scored 74: The approval surface is 7 operations and a request needs three fields. List endpoints take `limit`, and the MCP server's tool list could not be read without signing in (17 of 25). `limit` on 14 list operations, `cursor` or `offset`, `q`, status filters and eight audit filters (16 of 20). Errors carry a stable `type` and `code`, a `request_id` and structured `details`, with 409, 412 and 428 codes an agent can act on (18 of 20). Mandatory idempotency keys and `If-Match` on revision activation. MCP tool annotations were not checked (16 of 20). Sensible defaults and few required parameters, with hooks for six agent runtimes through the CLI, but no SDK package was found on npm or PyPI (7 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.

## 6. Transparency & trust, 53 out of 100, up to 4.1 more on the total

Made of editorial 52, provenance 53.

Why it scored 53: Closed hosted service with short, clear terms. The site calls the core open source, but the repository it links to returned 404 on 8 October 2026, so no licence could be read (15 of 30). The privacy policy and terms agree on no model training, server logs kept 30 days and records kept for the life of the organisation. The purge after deletion is only 'a short grace period', no DPA was found, and the AI Judge docs name OpenAI model keys while the processor list names no model provider (18 of 30). The terms promise notice of removals and 14 days' email notice of material changes to the terms, with no dated deprecation policy (5 of 20). Eight processors are named with purposes and hosting is stated as Azure UK South, less the missing model provider (14 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: withhuman.ai, registered 2026-07-13 (under a year) (0 of 15)
- Terms of service: read, states 6 of the 7 things a reader expects (9.1 of 10)
- Privacy policy: read, states 6 of the 8 things a reader expects (8.5 of 10)
- Status page: not found (0 of 10)
- Changelog: not found (0 of 10)
- security.txt: not found (0 of 10)

## 7. Schema & documentation, 81 out of 100, up to 3.1 more on the total

Why it scored 81: A public OpenAPI 3.1 document with 116 operations and 148 schemas, also served by each instance at `/api/openapi.yaml` (25). llms.txt lists every docs page with a Markdown copy (10). Operation descriptions state purpose, preconditions and what each credential may do, and the docs separate the MCP server from the MCP gateway (16 of 20). Enums for status, required fields and bounded durations are typed. `arguments`, `context` and instance metadata are free-form objects by design (12 of 15). Each operation has a curl example, sample responses and a response table, and the error envelope has a table of stable types (13 of 15). Paths are versioned (`/api/aap/v1`, `/api/v1`), but no changelog was found and the blog is empty (5 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.

## Deductions

Each comes off the total. A fixed and documented problem counts for less at the next check.

- 8 October 2026. The home page and pricing page describe an open-source, self-hostable edition and link https://github.com/withHumanAI/withHuman, which returned 404. The withHumanAI organisation listed no public repositories and the protocol repository https://github.com/agentapprovalprotocol/agentapprovalprotocol also returned 404. Counted as a claim the public cannot verify today (-2). Sources https://withhuman.ai/pricing and https://github.com/withHumanAI

## 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.

- The lead named an npm package `@withhuman/mcp` and SDKs. The npm registry returned 404 for `@withhuman/mcp`, `@withhuman/cli` and `@withhuman/sdk`, and PyPI for `withhuman`. The MCP server is hosted.
- Whether the open-source edition exists in public. Both linked repositories returned 404 on 8 October 2026, and the header shows the number 123 beside the GitHub link.
- unchecked: the MCP server's tool count, input schemas and annotations, which need a signed-in account.
- unchecked: whether signup or the trial asks for a card. The signup page is drawn by script and we did not create an account.
- unchecked: status.withhuman.ai did not connect through our proxy. No status page is linked anywhere on the site.
- Which company supplies the AI Judge models. The docs list keys such as `gpt-5.6-luna`, and the privacy policy's processor list names no model provider.
- What backs the SOC 2 Type II line in the site footer. No report, auditor or trust page was found.
- The numbers behind the Free plan's limits on pending approvals, gateway tool calls and phone messages.

## Weaknesses

- The terms say the service is in early access, and the API document is version 0.1.0
- No status page, published request rate limits, SLA outside an enterprise contract, or changelog found
- The GitHub repository the site links as its open-source edition returned 404 on 8 October 2026, and the organisation lists no public repositories
- No SDK found on npm or PyPI. Integration is the HTTP API, the CLI hooks or the MCP gateway
- The site footer says SOC 2 Type II with no report, auditor or trust page, and neither host serves a security.txt

## 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.

- Send an `Idempotency-Key` header on `POST /api/aap/v1/requests`. The call answers 400 without one
- Read the decision with `GET /api/aap/v1/requests/{id}?wait=30s` and repeat while `status` is `pending`. `wait` is capped at 30 seconds
- Set `timeout` between one second and seven days, and treat `expired` and `cancelled` as a refusal
- Start an approved call within five minutes of `decided_at`, because the decision carries an `expires_at`
- Cancel a request when you stop waiting, so reviewers stop seeing it

## 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.