Three unauthenticated render-only tools, and a shelf that is open self-serve on both platforms - the only real gates are Anthropic's AI-media clause and OpenAI's under-13 clause
The question
"What does the Goodnotes Claude/ChatGPT connector actually expose as a tool surface, and what did getting listed on those connector shelves require of Goodnotes?"
Promoted follow-up #1 of [[2026-10-08-goodnotes-generated-visuals-canvas]], which reframed the Goodnotes threat away from the canvas and onto the connector shelf. This brief answers both halves against primary sources, including the live Model Context Protocol (MCP) server itself.
How to read the labels. [DOC] is a documented fact traced to a source opened during this research. [OBS] is an observation from a live endpoint or page I queried directly. [INF] is my inference, which is not a finding. [UNKNOWN] means not publicly documented, and I stopped there rather than reasoning to a plausible answer.
What we already know (from the vault)
- The parent brief established the connector exists and bounded its output set, from one help-center article. [[2026-10-08-goodnotes-generated-visuals-canvas]] recorded "creates new documents only," "does not use your Goodnotes AI credits," "available on all Goodnotes plans," and dated the listing to April 2026 off a help-center search-index date. It explicitly flagged
https://www.anthropic.com/connectorsas HTTP 404 and the listing as "attested only by Goodnotes' own help article." - The product principle that constrains any RDCO version of this is already ratified. SW-R18 per [[2026-09-19-mentava-motivation-mechanics-paper-translation-scribble-works]]: parents on the site, kids on paper, no kid accounts, no kid profiles, no on-screen play, no third-party identity providers.
- The founder named this lane, after Goodnotes took it. [[2026-09-19-muse-connectors-agent-distribution]], founder by iMessage 2026-09-19: "If we are able to connect to agents that becomes an incredible distribution channel." Filed as RayDataCo/scribble-works#333, still unbuilt. That note also carries Isenberg's warning that "whoever you connect to becomes your landlord."
- Agent-originated calls already have a ruled-on billing shape. [[2026-08-31-product-model-rulings]]: agent calls (connector, MCP clients, any service credential) are rate-limited and budgeted on their own keys, separate from the human browser lane, but every call debits one household ledger.
- The adjacent directory brief covered the plugin path, not the connector path. [[2026-10-07-ray-plugins-marketplace-verified-eligibility]] closed the plugin pre-submission gate and concluded that "Anthropic Verified" is a connector label RDCO cannot earn with a skills-only bundle. It left the connector-side criteria open. This brief closes that side.
- The MCP substrate is mature enough that the shelf is a real channel. [[2026-07-28-mcp-spec-stateless-apps-tunnels]]: 950+ servers in the connectors directory, MCP Apps shipped, enterprise-managed auth available.
What the web says
1. The tool surface, read off the live server
The Claude directory listing publishes the connector endpoint, so the server's own tools/list is the primary source rather than any vendor prose. Both deployments answered an unauthenticated MCP handshake on 2026-10-20.
[OBS]The server is live, unauthenticated, and deployed twice - once per shelf.https://claude-mcp-api.ml.goodnotes.com/mcpandhttps://chatgpt-mcp-api.ml.goodnotes.com/mcpboth returnserverInfo: {"name":"goodnotes-mcp-server","version":"1.0.0"}over Streamable HTTP at protocol version2025-06-18, advertisecapabilities: {resources, tools}, and issue anmcp-session-id. No credential, token, or header was needed forinitialize,tools/list, orresources/list. The Claude directory listing statesSIGN-IN: Not requiredin its own metadata panel.[OBS]Exactly three tools, byte-identical on both platforms.create_markdown_text_document(requiredrawMarkdown,title),draw_svg_image(requiredtitle,svgCode; optionalbackgroundobject with literal colors white, black, yellow and patterns blank, dotted, squared), andgenerate_mermaid_diagram(requiredtitle,diagramTypeconstrained to flowchart, mindmap, timeline,mermaidCode; optionalthemewith colors classic, gray, lines and shades light, dark, plus the samebackground). Nothing else. Tool names, schemas, required fields, and descriptions matched exactly across the two hosts.[OBS]Every tool declares itself read-only. All three carryannotations: {"readOnlyHint": true, "destructiveHint": false, "openWorldHint": false}. Anthropic's own review criteria state that these annotations "determine auto-permissions in Claude. Read-only tools can run without per-call confirmation, and destructive tools always prompt." So the tool call itself is auto-permitted.[OBS]There is no read scope at all. No tool accepts a document identifier, a library query, a folder, a search string, or any account identifier. The only inputs are content the host model authored.resources/listreturns two user-interface widgets and no user data:ui://widget/whiteboard.html("Draw Notes Widget - SVG drawing widget template (basic shapes and text only)") andui://widget/markdown.html("Markdown Widget"), bothmimeType: text/html;profile=mcp-app.[DOC]The vendor's own stated limit matches that exactly. Goodnotes support: "The integration creates new documents only. It cannot edit or analyze documents already in your Goodnotes library," and "Each time you import generated or revised content from a chat, Goodnotes creates a separate new document in your library."[OBS]The metadata is platform-negotiated off the same tool contract. When I identified as Claude, each tool carried_meta.ui.resourceUriplus a_meta.ui.cspallowlist and a per-listing sandbox domain (37af3e0019f369343d0440d74e2f29cb.claudemcpcontent.com). When I re-ran the same call identifying as ChatGPT (clientInfo.name: chatgpt, headersx-mcp-clientandmcp-host), every tool gained a full OpenAI vocabulary:openai/outputTemplate,openai/widgetDomain,openai/widgetCSPwith separateconnect_domains,resource_domains, andredirect_domains,openai/widgetAccessible,openai/resultCanProduceWidget,openai/widgetPrefersBorder, and per-tool progress stringsopenai/toolInvocation/invoking: "Preparing your notes..."andinvoked: "Your notes are ready!". One tool contract, two platform dressings, selected by client identity.[OBS]The widget phones home to analytics. The content-security-policy allowlists on both platforms nameapi2.amplitude.com,events-webhook.insights.goodnotes.com,events-proxy.goodnotes-devops.com, andbrowser-intake-datadoghq.comalongside the Goodnotes web hosts.[UNKNOWN]what those events contain.[DOC]The human-confirmation boundary sits after generation, at a link, not at the tool call. Goodnotes support: "Review the generated preview," then "Select Edit in Goodnotes to import the content into your Goodnotes library," which "opens a link that imports the content into your Goodnotes library as a new document." It adds: "Make sure you are signed in to your Goodnotes Account and the Goodnotes app is open on your device. If the app is not open, the import may not complete successfully."[INF]The write into the user's library is client-side, which is why the server needs no OAuth. The server holds no account credential, has no read scope, and the import depends on the local app being open and signed in. That is consistent with a deep-link or web handoff executed by the already-authenticated Goodnotes client, not a server-to-server write. This also makesreadOnlyHint: truedefensible on tools whose names say "create": nothing in the user's account is mutated by the tool call. Label this an inference. I did not call any tool and did not observe the import.[DOC]Platform gating and cost. Goodnotes support: "This integration uses the AI model from ChatGPT or Claude and does not use your Goodnotes AI credits. It is available on all Goodnotes plans, including the Free plan. A Claude Pro, Team, or Enterprise plan may be required to use connectors."[OBS]The listing is Verified and predates the parent brief's date. The Claude directory entry athttps://claude.ai/directory/goodnotescarries a Verified badge in its accessibility labels, statesADDED March 2026, lists categories Files & documents, Designs & media, Notes & memory, and publishes the connector URL in the clear. March corrects the parent brief's April 2026, which came from a help-center search-index date rather than the listing.[DOC]Goodnotes publishes none of this itself outside one help article. Its/for-ai-assistantspage, stamped "Last Verified: September 2026," contains zero instances of "MCP," "connector," "ChatGPT," "Claude," or "OAuth," and/llms.txtmentions none of them either. The tool surface is discoverable from Anthropic's directory and the server, not from the vendor.
2. What listing required - Anthropic
Anthropic's connector documentation is first-party material about its own shelf, so treat the framing as promotional even where the requirements are concrete.
[DOC]The shelf is open self-serve, with no partner program. Submission runs through the developer portal atclaude.ai/directory/manage, choosing MCP connector. "Anyone on a paid Claude plan can submit, and on Team and Enterprise an Owner submits."[DOC]Technical conformance is MCP, remote, and HTTPS. The portal takes anhttps://URL. "A local server can't be submitted on its own." Local-server directory listings are deprecated and MCP Bundles are no longer accepted. The Software Directory Policy adds that "Remote MCP servers should support the Streamable HTTP transport. Servers may support SSE for the time being, but in the future it will be deprecated," and that tool names must not exceed 64 characters.[DOC]Authentication is conditional, and "none" is explicitly allowed. The checklist reads: "OAuth 2.0 if your tools act on a user's account, or no authentication for public data." The policy's binding form is narrower still: "Remote MCP servers that connect to a remote service and require authentication must use secure OAuth 2.0 with certificates from recognized authorities." Portal auth modes are OAuth with dynamic client registration, client ID metadata documents, Anthropic-held client credentials, a custom connection where users supply their own URL or credentials, or no authentication.[DOC]Tool-design rules are mechanical and are the most common rejection reason. Every tool needs atitleand the applicablereadOnlyHintordestructiveHint. "A single tool that accepts both safe HTTP methods, such as GET, HEAD, and OPTIONS, and unsafe methods, such as POST, PUT, PATCH, and DELETE, is rejected." Freeform query tools must name or link the target API. Descriptions must match behavior and must not instruct Claude how to behave, pull instructions from external sources, carry hidden or encoded instructions, or promote products.[DOC]Server behavior rules. Every tool must return a successful response on valid parameters; generic "Internal Server Error" responses fail review. Responses must be token-frugal. "Don't query Claude's memory, chat history, conversation summaries, or user files." The server must call your own first-party or legitimately proxied APIs, and "the MCP server domain should match your service."[DOC]Two use-case prohibitions, and one carve-out that matters to RDCO. Unsupported: software that transfers money or crypto, and "Software that uses AI models to generate images, video, or audio content." The carve-out, verbatim from the Software Directory Policy: "Design-focused software that uses AI models to create visual aids (such as slides, diagrams, charts, UI mockups, logos, or other design assets) are permitted. These servers may generate images as part of a design workflow, provided the developer does not offer standalone image generation as a primary service." Advertising and sponsored-content vehicles are also excluded.[DOC]Submission artifacts. Documentation URL, privacy policy URL, support contact, icon, 1-5 categories, a permanent URL slug, a name up to 100 characters, a one-liner up to 200, a description up to 2,000, company name and website, a primary review contact, a declaration of whether the connector reads data, writes data, or both, a data-handling declaration (own API, proxied with permission, or third-party), and test-account credentials "for a fully populated account." The policy adds that developers must provide "a standard testing account with sample data for Anthropic to verify full Software functionality" and "at least three working examples of prompts or use cases that demonstrate core functionality." Public documentation is "required by your publish date. A blog post or help-center article is sufficient."[DOC]Seven compliance acknowledgments, all required, covering directory guidelines, first-party application-programming-interface usage, financial transactions, AI media generation, prompt injection, conversation data collection, and public documentation.[DOC]No domain-ownership proof. "No domain-ownership proof (DNS or.well-known) is required. That requirement applies only to the open MCP Registry, not the Anthropic Directory."[DOC]Review is a scan by default, with automatic escalation to Verified. "Anthropic scans your submission automatically for policy compliance and, by default, lists it as a Community connector with no action from you." Anthropic "may then escalate listings flagged as highly useful to Claude users to Verified review, which is higher touch and slower, and in which reviewers run a functional test of each tool. That escalation is assessed automatically, and you don't need to take any action." Escalations go tomcp-review@anthropic.com.[DOC]Distribution mechanics that come with the listing. Every directory entry is automatically eligible for Suggested Connectors, in-chat recommendations shown when relevant to the user's task, and "ranking is usage-based, similar to other app stores." The catalog is shared across claude.ai, Cowork, Desktop, mobile, and Claude Code.[DOC]Legal instruments are the Anthropic Software Directory Terms and the Anthropic Software Directory Policy, plus commitments to maintain security and functionality, respond to security issues promptly, and provide accurate descriptions.[UNKNOWN]Review timeline. Anthropic states only that review times "vary with queue volume" and that review time "isn't fixed." No service-level commitment is published.[UNKNOWN]Fees, revenue share, or any commercial term. Nothing in the terms, policy, submission, or directory pages names a fee or a revenue share in either direction. Not publicly documented.
3. What listing required - OpenAI
Materially different, and the differences all run in the direction of more friction.
[DOC]The submission unit is no longer an "app," it is a plugin package. OpenAI's live documentation is titled "Plugin guidelines," governs "all plugins listed in the ChatGPT Directory," and describes a plugin as skills plus an optional MCP server plus optional user interface, uploaded as a ZIP to the portal atplatform.openai.com/plugins. Goodnotes' help article still calls it "an app in ChatGPT." "Your package can declare multiple MCP servers, but only one MCP server can be connected per plugin," and only one version may be published and one in review at a time.[DOC]Identity verification is mandatory and enforced at review. "Complete individual or business verification... Publishing under an unverified individual or business name will result in rejection." Submitting also requires theapi.apps.write/ "Apps Management Write" permission.[DOC]Domain verification is required. "For domain verification, host the exact challenge token as plain text at the URL shown in the portal." This is the single sharpest divergence from Anthropic, which explicitly waives it.[DOC]Server requirements. Publicly accessible domain, no local or testing endpoint, and if the server returns user interface, a content-security policy allowing the exact domains the component fetches from. Workspace-specific template URLs are gated: "We only support template-based URLs for trusted developers with whom we have an established relationship."[DOC]Scanning imports your annotations and you cannot argue around them. "Scan Tools" imports tool names, titles, descriptions, input and output schemas, security schemes,_metafields, tool annotations, linked UI resource metadata including CSP, and serverinstructions. "If your server advertisesreadOnlyHint: false, describing the tool as 'functionally read-only' in the justification doesn't make the tool read-only."[DOC]An age clause with no Anthropic equivalent. "Plugins must be suitable for general audiences, including users aged 13-17. Plugins may not explicitly target children under 13. Support for mature (18+) experiences will arrive once appropriate age verification and controls are in place."[DOC]Other content and commerce rules. No unofficial connectors or "pass-through intermediary software layers." No trial or demo plugins. Commerce is physical goods only: "Selling digital products or services - including subscriptions, digital content, tokens, or credits - is not allowed, whether offered directly or indirectly (for example, through freemium upsells)." A plugin may explain that a feature needs a different plan and may link to an informational plans page, but may not link to checkout. A published privacy policy is required covering categories of personal data, purposes, recipients, retention, and user controls. Restricted data that must not be collected: payment-card data, protected health information, government identifiers, and credentials.[DOC]Jurisdiction and residency constraints. Submit only if you intend public availability "in the countries you define during submission"; private use should go through a custom MCP server connection instead. "Projects with EU data residency cannot submit plugins with MCP servers for review."[DOC]Review process. Automated scans and/or manual review, status in the dashboard, email notification on status change, named feedback on failed checks, resubmission allowed, and appeal by replying to the rejection email. After first publication, MCP server changes are picked up automatically and eligible updates ship on passing automated checks, while metadata or skill changes need a new ZIP.[DOC]OpenAI publishes a conversion path from Anthropic. A guide titled "Submit your Claude Code plugin to OpenAI - Convert a Claude Code plugin or connector into a plugin you can submit to OpenAI" exists in the live documentation index.[UNKNOWN]Review timeline and any fee. No latency figure and no fee or revenue share is published. Not publicly documented.[UNKNOWN]The Goodnotes ChatGPT listing page itself. The ChatGPT plugin directory renders publicly while logged out, but its search and directory application-programming-interface return 401 Unauthorized without a session, and Goodnotes does not appear on the logged-out landing tiles. The ChatGPT-side listing is therefore attested by Goodnotes' help article naming the exact in-product flow, plus the live dedicatedchatgpt-mcp-apihost emitting a completeopenai/*widget contract. I did not open the listing page.
4. What neither platform documents about Goodnotes specifically
[UNKNOWN]When Goodnotes applied, to which shelf first, how long review took, who it talked to, and whether any partnership, marketing, or commercial agreement sat alongside the listing. Both directories are documented as self-serve with no partner program, and the listing shows only "ADDED March 2026" and a Verified badge. No application record, review correspondence, timeline, or agreement is publicly documented for Goodnotes. I am not going to reconstruct one.[UNKNOWN]Whether Goodnotes' Verified label came from the automatic escalation path or from any other route. Anthropic documents escalation as automatic with no separate application, which is consistent with what the listing shows, but the listing does not say how it was earned.- A note on the house Anthropic reference. The
claude-apiskill is not at~/.claude/skills/claude-api/; it is a bundled skill, which I loaded and searched. It documents the developer-side MCP connector (mcp_serversplus anmcp_toolsetentry under betamcp-client-2025-11-20) and contains no directory, shelf, or submission material. Anthropic's developer documentation index has zero instances of "directory," and its "Remote MCP servers" page renders its server list from a component rather than text, with the disclaimer that listed servers "are not owned, operated, or endorsed by Anthropic." The listing requirements live onclaude.com/docs/connectors/*and in the support-site policy, not in the developer docs.
Convergences and contradictions
- The parent brief's inference about the connector being deliberately thin is confirmed, and the mechanism is sharper than "deliberate." Goodnotes exposes three tools that accept model-authored markdown, Scalable Vector Graphics (SVG), and Mermaid source and render it.
[INF]That shape is what Anthropic's media clause permits: no image model runs server-side, so the connector is a design tool producing "diagrams, charts, or UI mockups" rather than software that "uses AI models to generate images." The narrowness may be a compliance artifact as much as a product choice. Inference, from the fit between the rule and the architecture, not from any Goodnotes statement. - The two platforms do not blend, and the asymmetry is one-directional. Anthropic: paid plan, portal form, annotations, OAuth only if needed, no domain proof, no identity verification, no country declaration, automatic scan, Community by default. OpenAI: identity verification enforced, domain-ownership challenge, a packaged ZIP, one MCP server per plugin, country declaration, no EU data residency, and an age clause. Every delta adds work on the OpenAI side.
- One contradiction inside Anthropic's own material. The connector pre-submission checklist renders the media prohibition flatly as "Generate images, video, or audio through AI models" with the single sentence "Design tools that produce diagrams, charts, or UI mockups are allowed." The binding Software Directory Policy is broader, naming "logos, or other design assets" and permitting image generation "as part of a design workflow, provided the developer does not offer standalone image generation as a primary service." A connector read against the checklist alone would look less eligible than it is. The policy is the instrument the checklist itself points to as "the full legal text."
- The shelf's content contradicts a strict reading of the media clause in practice.
[OBS]The logged-out ChatGPT plugin directory currently promotes "Magnific - Create images, video, designs" and "Higgsfield - Every image and video model." OpenAI has no equivalent prohibition, so that is consistent. Anthropic's own directory shows Adobe, Gamma, and Miro as related connectors on the Goodnotes page, all design-asset producers. The clause bites on standalone image generation as the primary service, not on visual output. - A fact the vault had wrong, now corrected. The listing date is March 2026, not April 2026. The parent brief's April came from a help-center index date. The gap between the shelf and the founder naming the lane on 2026-09-19 is six months, not five.
- The adjacent plugin brief's conclusion holds and gains a path. [[2026-10-07-ray-plugins-marketplace-verified-eligibility]] found Verified unreachable for a skills-only bundle. Goodnotes is the worked example of the route that does carry it: a remote MCP connector, submitted self-serve, escalated automatically, no application.
Synthesis for RDCO
The load-bearing finding is that the Goodnotes connector needs no authentication at all, and Anthropic's rules explicitly bless that. Three tools, all readOnlyHint: true, zero read scope, no account credential, no OAuth handshake, and a directory panel that says "SIGN-IN: Not required." The write into the user's library is handed off to an already-signed-in local client through a link the user taps. Read against SW-R18, this is not a constraint Scribble Works has to survive on the shelf; it is the compliant default. Anthropic's checklist offers "no authentication for public data" as a first-class option, and the policy only compels OAuth 2.0 for servers that "connect to a remote service and require authentication." A Scribble Works connector that exposes its catalog and returns a printable can be built with no accounts of any kind on either side of the call, which means no kid accounts, no kid profiles, no identity federation, and nothing to federate. The thing the founder's principle forbids turns out to be the thing the platform does not ask for.
That reframes scribble-works#333 from an access question to a content-policy question, and the two real gates are now named. The first is Anthropic's AI-media clause. Printable kids' activity sheets are image-bearing, and the exclusion of "Software that uses AI models to generate images, video, or audio content" is live. The carve-out permits "other design assets" generated "as part of a design workflow, provided the developer does not offer standalone image generation as a primary service," and Goodnotes shows exactly how to sit inside it: the host model authors SVG or Mermaid source, and the connector renders it deterministically. [INF] The transferable design is that a Scribble Works connector should render from its existing 81-game catalog and from model-authored layout, and should not call an image model at request time. If it does, the primary-service test gets hard to win; if it does not, the connector is a design tool with the same shape Anthropic already approved three times on one listing. That is my read, and it is a judgment about how a reviewer would apply a clause, not a documented ruling.
The second gate is OpenAI's, and it is the first real disqualifier found in either rulebook: "Plugins may not explicitly target children under 13." Scribble Works is explicitly for three-to-eight-year-olds. The clause is about who the plugin targets, and the honest ambiguity is whether a product whose user is a parent and whose subject is a four-year-old "explicitly targets" that child. [UNKNOWN] how OpenAI applies it; no worked example or policy elaboration is published. Against that, Anthropic's Software Directory Policy contains zero instances of "child," "minor," "under 13," "COPPA," "parent," or "teen" - I checked the full policy text, and the age dimension simply is not in Anthropic's rulebook. Combined with no domain verification, no identity verification, no ZIP package, no country declaration, and no EU residency restriction, the sequencing answer is not close: Anthropic first, and OpenAI only after the under-13 reading is resolved. OpenAI's own published conversion guide for Claude connectors makes that order cheap to follow later.
The channel economics point the same way the parent brief's cost analysis does, and that is the part worth carrying into the accepted-page math. Goodnotes' connector consumes the host's model tokens, not Goodnotes' own AI credits, and is free on every Goodnotes plan. A Scribble Works connector lane inherits the same split: the agent pays for authoring, RDCO pays only for rendering what it hands back. Set against the 5-to-8-cent target per accepted page in [[2026-10-01-accepted-spec-v0]], a connector lane where the model composes and RDCO renders is structurally cheaper per page than the site lane, and it arrives with [[2026-08-31-product-model-rulings]]' separate-key, one-ledger billing shape already ruled on. The listing also comes with distribution RDCO cannot buy: every directory entry is automatically eligible for in-chat Suggested Connectors, with usage-based ranking, across claude.ai, Cowork, Desktop, mobile, and Claude Code. [INF] That last point sharpens Isenberg's landlord warning rather than softening it, because usage-based ranking means the shelf rewards whoever is already being used, and a Verified badge is awarded by an automatic escalation RDCO cannot apply for.
Two product mechanics are directly liftable. First, the confirmation boundary. Because read-only tools auto-run, the user-facing gate is not the tool call but the widget's link out, and Anthropic provides an allowed-link-URI list that suppresses the "Open external link" prompt for origins the submitter owns. For Scribble Works that maps to: the tool returns a preview widget, the parent taps through to a PDF on scribbleworks.co, the parent stays the actor, and nothing is written to any account anywhere. Second, the dual-dressing trick. One MCP server, two hostnames, with platform-specific _meta selected off the client identity, is a single codebase serving both shelves. [OBS] I verified that by running the same tools/list twice with different client identity and watching the OpenAI widget vocabulary appear. That is the cheapest possible two-shelf architecture, and it is sitting in the open.
Why this is in the vault
It closes follow-up #1 of [[2026-10-08-goodnotes-generated-visuals-canvas]] with the connector's exact enumerated tool surface from the server itself rather than inference from a help article, and it converts RayDataCo/scribble-works#333 from "can we reach the shelf" into two specific go/no-go questions: whether a printable generator clears Anthropic's standalone-image-generation primary-service test, and whether Scribble Works clears OpenAI's under-13 clause. It also supplies the auth answer that SW-R18 was blocking on - no accounts are required by either platform for a generation-only connector - and it gives [[2026-10-07-ray-plugins-marketplace-verified-eligibility]] the connector-side path to the Verified label that the plugin-side path could not reach.
Open follow-ups
Each of these was checked against existing briefs in 06-reference/research/ via mcp__qmd__query on the rdco-vault collection before being filed, and each is a research question rather than a build task, a live-account test step, or something blocked on credentials.
- What shape of visual-output connector is actually live in the Anthropic directory today, read off the shelf rather than the clause? Adobe, Gamma, Miro, and Canva are all listed design-asset producers. Enumerating what each one's tools actually emit would turn the "standalone image generation as a primary service" test from a text reading into observed precedent, which is the single input that decides whether a Scribble Works connector is eligible. Not covered by [[2026-10-07-ray-plugins-marketplace-verified-eligibility]], which analysed the plugin scanner's policy prompt and not the media clause.
- How does OpenAI's "may not explicitly target children under 13" get applied in the published record? Any policy elaboration, developer-visible enforcement note, or rejection account would settle whether a parent-facing, child-subject product is in scope. This is the only clause found in either rulebook that could disqualify Scribble Works outright, and it currently rests on one sentence with no published gloss.
- What triggers an in-chat Suggested Connector? Anthropic documents that every directory entry is automatically eligible and that "ranking is usage-based," but not what makes Claude surface one mid-conversation. That mechanism is the entire distribution value of the listing, and nothing in the vault addresses it.
Related
- [[2026-10-08-goodnotes-generated-visuals-canvas]]
- [[2026-09-19-muse-connectors-agent-distribution]]
- [[2026-10-07-ray-plugins-marketplace-verified-eligibility]]
- [[2026-10-01-accepted-spec-v0]]
- [[2026-09-19-mentava-motivation-mechanics-paper-translation-scribble-works]]
- [[2026-08-31-product-model-rulings]]
- [[2026-07-28-mcp-spec-stateless-apps-tunnels]]
- [[2026-08-24-claude-marketplace-free-skill-paid-cta-policy]]
Sources
Vault
- [[2026-10-08-goodnotes-generated-visuals-canvas]] -
~/rdco-vault/06-reference/research/2026-10-08-goodnotes-generated-visuals-canvas.md(parent brief) - [[2026-10-07-ray-plugins-marketplace-verified-eligibility]] -
~/rdco-vault/06-reference/research/2026-10-07-ray-plugins-marketplace-verified-eligibility.md - [[2026-09-19-muse-connectors-agent-distribution]] -
~/rdco-vault/01-projects/printables-product/2026-09-19-muse-connectors-agent-distribution.md - [[2026-10-01-accepted-spec-v0]] -
~/rdco-vault/01-projects/printables-product/2026-10-01-accepted-spec-v0.md - [[2026-08-31-product-model-rulings]] -
~/rdco-vault/01-projects/printables-product/2026-08-31-product-model-rulings.md - [[2026-09-19-mentava-motivation-mechanics-paper-translation-scribble-works]] -
~/rdco-vault/06-reference/research/2026-09-19-mentava-motivation-mechanics-paper-translation-scribble-works.md - [[2026-07-28-mcp-spec-stateless-apps-tunnels]] -
~/rdco-vault/06-reference/2026-07-28-mcp-spec-stateless-apps-tunnels.md - [[2026-08-24-claude-marketplace-free-skill-paid-cta-policy]] -
~/rdco-vault/06-reference/research/2026-08-24-claude-marketplace-free-skill-paid-cta-policy.md
Live endpoints queried 2026-10-20 (read-only MCP metadata calls: initialize, tools/list, resources/list; no tool was invoked)
https://claude-mcp-api.ml.goodnotes.com/mcp- three tools, annotations, input schemas, two MCP App widget resources, Claude-side_meta.uiand CSPhttps://chatgpt-mcp-api.ml.goodnotes.com/mcp- identical tool contract; fullopenai/*widget metadata returned when the client identifies as ChatGPTmcp.goodnotes.com- does not resolve in DNS. There is no public MCP hostname under that name.
Web - Anthropic first-party, fetched 2026-10-20. Vendor documentation about its own shelf is promotional material, not neutral.
- https://claude.ai/directory/goodnotes - the listing: three tool names,
SIGN-IN: Not required, connector URL,ADDED March 2026, Verified badge, categories. Reached via Playwright;/directory/connectors/goodnotesredirects here - https://claude.com/docs/connectors/building/submission - portal steps, requirements, allowed link URIs, seven compliance acknowledgments,
mcp-review@anthropic.com - https://claude.com/docs/connectors/building/review-criteria - read/write split, annotations, 64-character names, prompt-injection rejections, unsupported use cases, Community and Verified review
- https://claude.com/docs/connectors/directory - self-serve submission, Suggested Connectors, usage-based ranking, "No domain-ownership proof... is required"
- https://support.claude.com/en/articles/13145358-anthropic-software-directory-policy - the binding instrument: section 4 unsupported use cases with the design-asset carve-out, section 3.D testing account, 3.E three working examples, section 5 MCP requirements (Streamable HTTP, OAuth 2.0 with recognized certificates, annotations, 64 characters)
- https://platform.claude.com/docs/en/agents-and-tools/remote-mcp-servers.md - third-party server list rendered from a component; "not owned, operated, or endorsed by Anthropic"
- https://platform.claude.com/llms.txt - zero instances of "directory"; the listing process is not in the developer docs
claude-apibundled skill (/private/tmp/claude-501/bundled-skills/2.1.304/.../claude-api) - house Anthropic reference; documents the API-side MCP connector client only, no listing material
Web - OpenAI first-party, fetched 2026-10-20. Same promotional caveat.
- https://developers.openai.com/plugins/plugin-guidelines.md - ChatGPT Directory requirements: appropriateness and the under-13 clause, unofficial-connector ban, commerce rules, privacy policy minimums, restricted data
- https://developers.openai.com/plugins/deploy/app-review.md - organization and individual identity verification,
api.apps.write, server requirements, template-URL trust gate, Scan Tools metadata import, annotation-justification rule, EU data residency exclusion, review and appeals - https://developers.openai.com/plugins/deploy/submission.md - ZIP package flow, publishing identity, domain-verification challenge token, one MCP server per plugin, automatic pickup of server changes after publication
- https://developers.openai.com/plugins/llms.txt - documentation map, including "Submit your Claude Code plugin to OpenAI"
- https://chatgpt.com/plugins - public directory landing, browsed logged out via Playwright; shows AI image and video generators on the shelf
Web - Goodnotes first-party, fetched 2026-10-20
- https://support.goodnotes.com/hc/en-us/articles/15790373823887-Use-Goodnotes-in-ChatGPT-and-Claude - full article via Playwright (curl and WebFetch both HTTP 403): "available as an app in ChatGPT and as a connector in Claude," the per-platform setup flows, the three content types, "Edit in Goodnotes," the signed-in-and-app-open requirement, "creates new documents only," credits and plan statements
- https://www.goodnotes.com/for-ai-assistants - zero mentions of MCP, connector, ChatGPT, Claude, or OAuth
- https://www.goodnotes.com/llms.txt - same absence
Flagged - not usable as fact
- The ChatGPT directory listing page for Goodnotes.
chatgpt.com/pluginsrenders logged out, but its search and directory application-programming-interface return 401 Unauthorized without a session, and Goodnotes is not on the logged-out landing tiles. Not retried behind a login. The ChatGPT-side listing is attested by the Goodnotes help article and the livechatgpt-mcp-apihost, not by the listing page. https://www.claude.com/directoryandhttps://claude.com/directory- HTTP 403 to curl. The working surface isclaude.ai/directory.- A WebSearch result digest was used only to locate the
claude.com/docs/connectors/*pages. Every requirement above was then re-read from the page or policy text itself; no claim in this brief rests on the digest. - Goodnotes' application date, review duration, counterparties, and any commercial or partnership terms. Not publicly documented on either platform, and not inferred here.