# Fix list: Subframe

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/subframe, the October 2026 research run, assessed 9 October 2026. Grade E, 39.7 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 Subframe: 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, 19 out of 100, up to 16.2 more on the total

Why it scored 19: Read with the hosted lines, because the surface an agent uses is the MCP server at mcp.subframe.com. No status page was found on the site, the docs or the footer (0 of 20). With no page there is no incident history to read (5 of 30). No rate limit is published in the docs or the skill files (0 of 15). No 429 or backoff guidance was found. The docs do say a background job that stops reporting for about 10 minutes is returned as `error`, and the CLI retries a failed request once (4 of 15). No SLA is published, and the terms say the service is not warranted to be uninterrupted (0 of 10). The MCP server carries no beta label, though the CLI's `push-component` command is marked experimental (10 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. Security & auth, 44 out of 100, up to 9.8 more on the total

Why it scored 44: The MCP server uses OAuth with dynamic client registration, PKCE and refresh tokens, and rejects static tokens. The authorisation server lists only identity scopes (openid, profile, email, phone, offline_access), so access follows the user's team role and not a scope. CLI tokens are shown once and can be deleted, with no scopes or expiry documented, and the `generate_auth_token` tool lets a connected agent mint one (22 of 30). Viewers get a read-only MCP server and read-only CLI access. Confirmation before deletes is guidance in the skill file and not a documented server control (12 of 20). `list_comments`, design documents and `search_docs` return text other people wrote, and no prompt-injection guidance was found (3 of 15). No audit log is documented. Version history records changes for 24 hours on Free and 7 days on Pro (3 of 15). No security.txt, disclosure policy, bug bounty or certification was found. The DPA's Annex 2 lists security measures and the docs cover Okta SSO (4 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.

## 3. Agent ergonomics, 47 out of 100, up to 8.6 more on the total

Why it scored 47: 46 documented tools is over 30 (5), with 3 added back because a Viewer account gets a read-only server. No toolsets or dynamic loading are documented, and the `design` skill adds about 55 KB when loaded (8 of 25). `list_comments` can be scoped to a page, a canvas or the project, screenshots can target one element or breakpoint and `read_prototype_file` returns one file. No paging is documented (10 of 20). `wait_for_jobs` reports `running`, `done`, `error` or `not_found`, delete tools refuse when the item is referenced, `edit_page` returns `appliedCode` and parser warnings, and the CLI prints a JSON error envelope (13 of 20). No idempotency keys. Tool annotations could not be read. Deletes need `force` when references exist, and version history can restore (8 of 20). `projectId` falls back to the first project and items are addressed by id, name or URL. No API SDK (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. Payments & pricing, 33 out of 100, up to 8.4 more on the total

Why it scored 33: No machine payment protocol (0 of 40). Plan prices are public, Free at $0 and Pro at $20 an editor a month, with AI credits described only as limited and 6x and no per-call price (10 of 20). The Free plan has one project and includes MCP and CLI access, and the docs mention a card only at the Stripe checkout for Pro (20 of 20). A person has to sign in through a browser to approve OAuth. A client can register itself with the authorisation server and a connected agent can mint a CLI token with `generate_auth_token`, which earns 3 (3 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.

## 5. Schema & documentation, 54 out of 100, up to 7.5 more on the total

Why it scored 54: The MCP server is closed and answers 401 without OAuth, so the tool definitions and their input schemas were not read. The docs list 46 tools by name and the public skill files give parameters for most of them (12 of 25, on that partial evidence). docs.subframe.com publishes `llms.txt`, a Markdown twin of each page and a docs MCP server (10 of 10). The `design` skill states for each tool what it is for and when another tool fits better, for example `edit_page` against `design_page`. The docs table gives one line a tool (15 of 20). Parameters are named in the skill (`id`, `name` or `url`, `force`, `deleteChildPages`, `includeResolved`), while the design tools take a free-text description (6 of 15). Ten example prompts and a troubleshooting section, with no list of error responses (7 of 15). The plugin and the CLI carry version numbers, with no changelog (4 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.

## 6. Transparency & trust, 57 out of 100, up to 3.8 more on the total

Made of editorial 46, provenance 68.

Why it scored 57: The service is closed under terms of service dated 22 January 2026. The CLI and `@subframe/core` are public and marked ISC in their package files, and the plugin MIT in its manifest, but the repository has no licence file (17 of 30). Terms, privacy policy (22 January 2026) and DPA (21 January 2026) agree with each other. The privacy policy gives no retention periods, and the terms say 'Unless Customer is notified otherwise, Customer Data is not used to train AI models', which leaves the default open to change by notice (15 of 30). No deprecation policy. A migration page explains the move to component directories, and three MCP tools were renamed with no notice found (4 of 20). The DPA links a sub-processor list held in a Google Sheet, which we did not read. The privacy policy names Google Analytics, PostHog, Segment and LogRocket, and the CLI's README discloses crash and usage reporting with `DO_NOT_TRACK=1` as the opt-out (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):

- Terms of service: read, states 5 of the 7 things a reader expects (8.3 of 10)
- Status page: not found (0 of 10)
- Changelog: not found (0 of 10)
- security.txt: not found (0 of 10)

## 7. Maintenance & community, 65 out of 100, up to 3.1 more on the total

Why it scored 65: The Claude Code plugin reached 1.0.30 on 8 October 2026 and `@subframe/cli` 1.212.0 on 5 October 2026 (30 of 30). The plugin version changed ten times between 13 July and 8 October, and the CLI had three versions on 5 October after none since 24 June (20 of 20). The repository has 13 issues in all. Seven are open, among them a Codex MCP install error from 30 March 2026 and a Claude Code plugin install error from 13 February 2026 with one and two comments, and three with none. Pull requests merge most days. The Slack community was not read (10 of 25). A search of the official MCP registry for 'subframe' returned no entry (0 of 15). A commit of 12 August 2026 fixed dependency vulnerabilities. The CLI has unit and end-to-end tests in the repository, and the only public workflow publishes the package (5 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.

## Deductions

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

- 23 to 24 September 2026. The docs and skill files renamed three MCP tools (`list_flows`, `get_flow_info`, `delete_flow`) to `list_canvases`, `get_canvas_info` and `delete_canvas`, with no changelog entry or notice found. Whether the server still answers the old names was not tested (-2). https://github.com/SubframeApp/subframe/commit/43bfb51d749713940cf0df8e73da2ab43b7358bc
- 7 May 2026. A docs commit records that the MCP server 'no longer accepts CLI/access tokens, only OAuth', removing the documented header route for clients without OAuth, with no dated notice found (-1, older and documented since). https://github.com/SubframeApp/subframe/commit/ff59743593fa2b581f32233b58660ace8a23f707

## 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 MCP tool definitions. mcp.subframe.com answers 401 without OAuth and the server is closed, so input schemas, descriptions and annotations were read only from the docs and the public skill files.
- unchecked: the sub-processor list, which the DPA links as a Google Sheet.
- unchecked: the Slack community, so support responsiveness there is unknown.
- Whether the server still accepts the three old flow tool names after the September 2026 rename.
- Whether sign-up asks for a card or a particular sign-in method. The docs mention a card only at Pro checkout.
- No status page, changelog, rate limit, SLA or security contact was found on the site, the docs or the policies site. status and changelog addresses were not guessed.
- The amount of AI credit on each plan is not published (Free 'Limited', Pro '6x').
- The lead called the CLI open source. Its source is public and its package file says ISC, but the repository has no licence file and GitHub reports no licence.
- The `design` skill's front matter tells a model to load it for any action through the Subframe MCP server. We read it as data.
- No first release date was established. `@subframe/cli` was created on npm on 17 May 2023.

## Weaknesses

- 46 tools are documented with no toolsets, and the vendor's `design` skill that explains them is about 55 KB of text
- No status page, published rate limit, SLA or changelog was found, and the terms disclaim uninterrupted service
- `list_flows`, `get_flow_info` and `delete_flow` became `list_canvases`, `get_canvas_info` and `delete_canvas` in September 2026 with no notice found
- No security.txt, disclosure policy, bug bounty, certification or audit log was found
- Delete confirmation is guidance in the skill file. The vendor's skill calls deletes irreversible from MCP, with recovery only through version history in the editor (24 hours on Free)

## 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.

- Connect with an MCP client that supports OAuth. The server rejects Subframe access tokens and static `Authorization` headers
- Pass `projectId` on every call. When it is omitted the server uses the first project the user can reach, which may be the wrong team's
- After `design_page`, `design_component` or `edit_component`, call `wait_for_jobs` with the `jobId` before reading the result. Earlier reads return stale content
- Ask the owner before any delete tool or a theme token deletion. A deleted token leaves every reference detached even after the token is restored
- For the CLI, set `SUBFRAME_AUTH_TOKEN` and `DO_NOT_TRACK=1`, and point `--dir` at a folder that holds only Subframe code, because a full sync removes other unprotected files

## 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.