# MCP directory listing steps kit

Checked November 7, 2026. For a small business owner and the person maintaining its website.

Companion: [Where to List Your MCP Connector in 2026, Step by Step](https://jwatte.com/blog/mcp-directories-listing-guide-2026/).

Use this worksheet to prepare your own server, publish accurate entries, and measure enquiries. Replace bracketed text and example.com addresses. These are proposed templates. Never copy a privacy promise without checking your implementation. Publishing and testing steps below are instructions for your authorized team; this kit does not submit anything.

MCP means Model Context Protocol, a shared way for an assistant to ask software for information or actions. An MCP server is the service answering those requests. A tool is one named operation, such as checking opening hours. A connector links an assistant to that server. A plugin can package a connector with reusable instructions called skills. A registry holds machine-readable records; a directory helps people find and connect services. An endpoint is the server's connection address. An API is an interface software uses to request information. A repository is a shared project folder with change history; a manifest describes a package, and a README explains setup. A pull request proposes a repository change for review. A CLI is a terminal program.

## 1. Assign the work

| Responsibility | Person or role | Done when |
|---|---|---|
| Business facts and claims | [owner] | Examples, limits and support promises match operations |
| Server and security | [developer] | Own-server tests pass and findings are resolved |
| Documentation and privacy | [web person plus owner] | Public pages match actual data handling |
| Submission and listings | [owner] | Actual public entry is visible and checked |
| Measurement and updates | [named role] | Monthly counts and release checks are recorded |

Business display name: [name]
Public server: https://example.com/mcp
Documentation: https://example.com/connect/
Privacy: https://example.com/privacy/#mcp
Support: https://example.com/support/
Terms: https://example.com/terms/
Current release: [version]
Customer task: [one useful task]
Human decision: [what the person must review and submit]

## 2. Readiness checklist before any directory

- [ ] Our own server answers over HTTPS. Record UTC time, negotiated protocol, server name and version from initialize.
- [ ] tools/list returns the intended tools. Compare every name, title, input and effect with the documentation.
- [ ] Each tool has a title and applicable readOnlyHint or destructiveHint. A read-only hint declares that an operation only reads; it does not enforce permissions. A destructive hint describes potentially destructive changes. OpenAI requires explicit boolean values for all three effect annotations, including openWorldHint. Its current plugin guidelines say annotation justifications are no longer required.
- [ ] Each description says what the tool does, required inputs, side effects and what it cannot do. A returned form link must not claim a message was sent.
- [ ] Errors distinguish unavailable options, invalid inputs, sign-in failures and unsupported requests, with a useful next step.
- [ ] The human documentation page explains connection, prompts, limits and support. Our [connect page](https://jwatte.com/connect/) is one public example.
- [ ] The privacy section describes inputs, successful and failed request logs, hosting providers, retention and any sharing. Our [MCP privacy section](https://jwatte.com/privacy/#mcp) shows the placement pattern.
- [ ] The owner has confirmed the support contact receives mail through an authorized mailbox test. A published address alone does not establish delivery.
- [ ] Run a pre-release security check of your own server code. [MCP-Sentinel](https://github.com/BashaarJavaid/MCP-Sentinel) is one open example, licensed MIT according to its [GitHub license record](https://api.github.com/repos/BashaarJavaid/MCP-Sentinel/license), read November 7 at 18:43:04 UTC. We did not install or run it. See the [companion server-security guidance](https://jwatte.com/blog/build-mcp-connector-nonprofits-events-wordpress/).
- [ ] Test expected results, unavailable choices, ambiguous inputs and refusals in the intended assistant using your own server. Do not create live orders or send messages merely to test a read-only operation.
- [ ] Owner checks any publication declarations personally against evidence.

Source requirements: [Claude readiness](https://claude.com/docs/connectors/building/submission.md), read November 7 at 18:42:59 UTC; [OpenAI submission validation](https://developers.openai.com/plugins/deploy/submission-errors), read 18:42:59 UTC.

Use [MCP Server Audit](https://jwatte.com/tools/mcp-server-audit/) for public-file and endpoint-address GET reads; it does not initialize or execute tools. [Agent Endpoint Inspector](https://jwatte.com/tools/agent-endpoint-inspector/) starts with passive reads and offers protocol checks only after the owner explicitly selects the handshake option. Do not treat a scanner result as directory approval or permission to test someone else's server.

Test record:

| Check | Expected result | Observed result and UTC time | Responsible role |
|---|---|---|---|
| initialize | [version and protocol] | [result] | [developer] |
| tools/list | [tools and annotations] | [result] | [developer] |
| Normal customer request | [answer and own-page link] | [result] | [owner] |
| Unsupported request | [clear refusal] | [result] | [owner] |
| Link handoff | [prefill, no submission] | [result] | [web person] |
| Privacy and support | [accurate explanation and working route] | [result] | [owner] |

## 3. Public record card: Official MCP Registry

Needs: developer, remote HTTPS server, server.json, namespace ownership proof and release version. server.json is the structured description other software reads. A namespace is the publisher's name prefix: example.com becomes com.example. A DNS TXT record is a short public text record managed with your domain's settings.

1. Read the [remote-server instructions](https://modelcontextprotocol.io/registry/remote-servers) and [publisher quickstart](https://modelcontextprotocol.io/registry/quickstart). The quickstart's package example is not a requirement to publish a package for a remote-only server.
2. Prepare server.json with name, description, version and remotes entry pointing to your HTTPS endpoint with streamable-http transport. Use the current official schema.
3. Follow [authentication instructions](https://modelcontextprotocol.io/registry/authentication) for reverse-domain proof using DNS TXT or a public file at /.well-known/mcp-registry-auth. Keep signing keys private. GitHub authentication instead uses an io.github namespace.
4. Have your authorized developer install mcp-publisher from the official instructions, create and validate the metadata, authenticate for the intended namespace, and publish. Those are system-changing and publication actions; review the finished entry first.
5. Read your record through the v0.1 API. Check name, remote URL, version, active status and latest-version metadata.

Original read-only shell example: replace the example name with your own namespace and server. The encoded %2F represents the slash between them. This reads Registry metadata and does not call your MCP server.

```sh
curl -fSs 'https://registry.modelcontextprotocol.io/v0.1/servers/com.example%2Fpublic-information/versions/latest'
```

Validation: the shell example was parsed with bash -n. It was not run against the placeholder name. Our own record was separately read at [jwatte's latest Registry entry](https://registry.modelcontextprotocol.io/v0.1/servers/com.jwatte%2Fsite/versions/latest) on November 7, 2026; our six active/latest records were checked between 18:43:34 and 18:43:41 UTC.

Confirm: save the actual Registry record URL in the tracker. Update: publish a new unique version; published metadata cannot be overwritten. Metrics: the Registry is a metadata feed, and these docs establish no owner lead dashboard. It performs no security review. Independent services decide what to import and when. Read [versioning](https://modelcontextprotocol.io/registry/versioning) and [Registry scope](https://modelcontextprotocol.io/registry/about). These pages were read November 7 at 18:43:04 and 18:44:30 UTC; publishing/authentication pages at 18:42:58 to 18:43:01 UTC.

## 4. Assistant directory cards

### Claude connector

Needs: paid submitting account with the appropriate organization permissions, own HTTPS server, icon, public documentation/privacy/support, accurate authentication and reviewer instructions. OAuth is the sign-in permission flow that lets a user authorize an app without giving it their password. Public information can require no sign-in.

1. Test your own server as a custom connector, meaning a connection added directly by its address.
2. Open [developer portal](https://claude.ai/directory/manage), choose Submit new, then MCP connector.
3. Connect the server, fix detected tool warnings, and complete listing details, use cases, company, authentication and data handling.
4. Supply reviewer access and test instructions, review declarations, and submit. The [ten-stage walkthrough](https://jwatte.com/blog/mcp-connector-listing-small-business/) has form-level detail.
5. Follow the portal status and feedback through publication; check the actual public entry and repeat your customer prompts.

Confirm: the public listing is visible with correct tools and sign-in wording. Update: deploy server changes and edit copied listing details in the portal; listing edits may await review. The slug stays fixed. Metrics: connector Overview/Usage reports activity, health and errors. Figures update daily, can lag up to 24 hours and omit low-volume results. Empty charts do not prove nobody tried it.

Community screening is limited review. Verified adds closer review and functional testing, but is not a security audit. There is no separate Verified application. No fixed review time is documented. Sources: [submission](https://claude.com/docs/connectors/building/submission.md), [verification](https://claude.com/docs/connectors/verification.md), [metrics](https://claude.com/docs/connectors/building/managing-your-listing.md), read November 7 at 18:42:59, 18:43:00 and 18:43:01 UTC.

### Claude plugin bundle and Marketplace distinction

Needs: a GitHub-hosted bundle with supported components, public repository before publication, correct owning organization and tested instructions.

1. Use [publish documentation](https://claude.com/docs/directory/publish) to choose Plugin bundle in the same portal.
2. Prepare the bundle and submit its repository through that workflow. If you operate a remote server it references, submit that separately as an MCP connector under the same organization, then pair them.
3. Resolve validation and review findings; verify the published bundle on supported Claude surfaces.

Confirm: the intended bundle and connector listings are visible. Update: publish to the tracked repository branch and check the scanned version in the portal. Metrics: plugin Usage includes installs, versions, component use and a listing-to-install funnel; until usage arrives, Preview can show sample figures. [Current publishing and update docs](https://claude.com/docs/connectors/building/after-publishing.md), read November 7 at 18:43:03 UTC.

Claude Marketplace is also a website for discovering partner products and services. Connector/plugin submissions use the directory route; selling a Claude-powered product or joining as a service partner has a different path. [September 23 Marketplace announcement](https://claude.com/resources/articles/claude-marketplace), read November 7 at 18:42:31 UTC; [September 25 portal announcement](https://claude.com/resources/articles/build-plugins-for-claude), read 18:42:28 UTC. No enterprise purchasing access follows automatically from listing your server.

### ChatGPT personal test

Needs: account/workspace access to custom MCP connections and an authorized server test plan.

1. In [ChatGPT Plugins](https://chatgpt.com/plugins), choose the plus control and Add custom MCP server.
2. Enter name, description and public HTTPS server URL, configure sign-in, review the risks and create the plugin.
3. Inspect discovered tools, install the resulting plugin, start a new chat and select it with @.
4. Run success and refusal tests. Record selected tool, arguments, results and any confirmation.

Confirm: useful tool execution is recorded, not just an added connection. Update: Refresh the custom connection's metadata, then retest in a new chat. Metrics: keep your own controlled test record; this is not public distribution. [OpenAI connection guide](https://developers.openai.com/plugins/deploy/connect-chatgpt), read November 7 at 18:43:03 UTC. Account and workspace policy determine access.

### ChatGPT public plugin

Needs: verified individual or business publishing identity, organization permissions, plugin ZIP, production HTTPS endpoint, accurate annotations, website/support/privacy/terms URLs, review cases and a video.

1. Read the [current submission guide](https://developers.openai.com/plugins/deploy/submission) and [plugin guidelines](https://developers.openai.com/plugins/plugin-guidelines). Choose the organization/project that should own it. Prepare the MCP-backed package from the start.
2. Use the portal's With MCP route and upload the ZIP with your verified publishing identity. Keep credentials out of the package.
3. Resolve metadata and skill findings. Connect the server and serve the exact plain-text domain challenge at the portal's requested /.well-known/openai-apps-challenge address. This is a single token, not JSON.
4. Inspect tool scans. Supply explicit readOnlyHint, destructiveHint and openWorldHint. Current plugin guidelines say annotation justifications are no longer required, although older review wording and validation documentation still mention them. Follow the current guidelines and resolve any portal findings. Correct the server and rescan as needed.
5. Provide five positive and three negative cases, accessible video walkthrough and release notes. Supply dedicated reviewer credentials separately when sign-in is required, with sample data and no inaccessible approval or code step.
6. Submit for review, resolve findings, then select Publish plugin when approved. Verify the live entry and customer experience.

Confirm: approved package is actually published. Update: hosted tools are scanned daily; eligible changes become available after checks. Use Rescan after changes and preserve compatibility with approved definitions while updates wait. Metadata or skills changes need a new ZIP. Metrics: these submission pages establish review/scan status, not a promised owner lead dashboard. Review completion time was not established. Skill scans can take up to two hours; that is not a review deadline. [Submission validation](https://developers.openai.com/plugins/deploy/submission-errors), read November 7 at 18:42:59 UTC. Submission and guidelines read at 18:42:58 and 18:43:01 UTC.

Do not make the plugin primarily an advertisement. Give it useful standalone functionality. Follow current restricted-input and commerce rules before proposing purchases or collecting customer details; this worksheet does not establish eligibility.

### Gemini developer test

Needs: own HTTPS MCP server and an account offering Developer Mode.

1. Open Settings > Gemini Apps Developer Mode, or [Developer Mode](https://gemini.google.com/devmode).
2. Choose Create new, enter the endpoint and configure authentication where needed.
3. Review the security/privacy screen. Name the custom app and confirm Available actions.
4. Try natural-language and @mention requests. Inspect Tool schemas and Event Log to verify actual calls and expected confirmations.

Confirm: requests execute correctly in the test workspace. Update: Sync schema and retest. Metrics: Event Log is debugging evidence, not directory reach. [Google testing instructions](https://developers.google.com/connected-apps/mcp/dev-mode/usage), read November 7 at 18:43:04 UTC.

### Google Connected Apps preview

Needs: preview admission, authorized developer account, public HTTPS server, listing assets, domain control, terms/privacy pages and reviewer test data where relevant.

1. Apply from [Connected Apps Platform](https://developers.google.com/connected-apps). A private Gemini test does not confer admission.
2. In Connected Console create the app, select category and MCP server connection type, and complete public details and prompts.
3. Host the console's JSON domain challenge at /.well-known/google-connected-apps, returning 200 without redirects. Configure no authentication for a public service or the appropriate authorization flow.
4. Sync tools into a version snapshot, resolve errors, select countries/products/languages and confirmation requirements. Add testers and test the entire flow. Google documents up to ten minutes for tester visibility.
5. Send for review. Address issues and select Publish after approval.

Confirm: Published status and availability in the selected supported product. Update: make a new version, resync tools, test, review and publish. Metrics: the platform describes usage analysis, but exact production reporting fields were not established in these pages. No review-duration promise is given. [Publication steps](https://developers.google.com/connected-apps/gpc/perform-the-steps) and [prerequisites](https://developers.google.com/connected-apps/gpc/prerequisites), read November 7 at 18:43:05 and 18:43:07 UTC; overview read 18:42:34 UTC.

## 5. Independent index cards

### Glama

Needs: accurate Registry entry or repository/hosted endpoint details; proof of control if claiming; dedicated test access when authentication is required.

1. Publish Registry metadata, then search Glama because it documents Registry intake.
2. If absent, follow Add MCP Server for a repository or Add Server > Hosted endpoint for a hosted URL.
3. Claim the entry using the offered endpoint, domain or GitHub proof. Set up an appropriate test profile for authenticated checks.
4. Compare public name, URL, sign-in label, tools and findings with your documentation.

Confirm: correct public entry. Update: Registry controls name, description and URL by default; claimed owners can choose Glama listing details instead. Record that choice. Metrics: claimed owners receive listing management, health checks and analytics; these are not proof of enquiries. Most GitHub submissions are described as passing within minutes; no equivalent hosted-server timing promise was established. Ownership badges prove control, not correct business outcomes. [FAQ](https://glama.ai/mcp/faq), read November 7 at 18:42:24 UTC; [methodology](https://glama.ai/mcp/methodology), read 18:43:27 UTC.

### mcptry

Needs: Registry metadata with a remote URL, or the direct hosted-server submission details.

1. Publish the Registry entry and search mcptry after its documented approximate thirty-minute intake interval.
2. Review the entry's endpoint and tool-text findings.
3. Optionally claim domain/GitHub control. Claiming is free and is not required to remain listed.
4. If the hosted server is absent from Registry, the direct route uses GitHub sign-in and a quick review; a later Registry entry with the same URL replaces it.

Confirm: actual listing. Update: correct Registry data and check its next imported version. Metrics: public health/recency findings are established; an owner conversion dashboard was not established. Its verified badge indicates publisher identity, and methodology says payment cannot buy higher rank. Its scheduled checks do not execute tools, but its playground can. [Submission](https://mcptry.com/submit), read November 7 at 18:42:24 UTC; [methodology](https://mcptry.com/methodology), read 18:43:27 UTC.

### PulseMCP

Needs: accurate Registry metadata and a tracker entry marked paused.

1. Read the [current submission notice](https://www.pulsemcp.com/submit).
2. Publish or update the official Registry record now.
3. Watch the notice and verify the actual entry after intake resumes.

Confirm: future public listing, not merely Registry publication. Update: both new submissions and edits are currently paused. PulseMCP says it will import Registry listings once back; no reopening date is promised. Metrics/claim procedure: not established by this notice. Read in a browser November 7 at 18:44:05 UTC. The notice remains dated September 3, 2026.

### mcp.so

Needs: remote endpoint, name and a considered decision about a paid listing.

1. Open the [Remote Server form](https://mcp.so/submit?type=remote-server).
2. Prepare endpoint and name; confirm correction/ownership arrangements before paying.
3. If the owner chooses this paid route, inspect the resulting public entry against those details.

Confirm: correct listing after submission. Update/claim/owner metrics: not documented on the form checked. Its paid Verified badge accompanies a promise of immediate publication without review; it does not establish security review. No price is reproduced here. Browser read November 7 at 18:44:12 UTC.

### MCP Market

Needs: repository or remote endpoint, contact email, decision about free queue versus paid placement.

1. Choose MCP Server, then GitHub repository or Remote MCP on the [submission page](https://mcpmarket.com/submit).
2. Supply the requested details and contact email.
3. Choose the route deliberately and retain the confirmation.
4. Check the public entry when notified.

Confirm: actual published entry. Update/claim/metrics: not documented in the form, so establish the correction route before depending on it. The form advertises an average four to six weeks for free and within twenty-four hours for paid; these are the vendor's statements, not measured waiting times. The paid Official badge is not assistant approval or proof of domain control. Read November 7 at 18:42:24 UTC.

### Smithery

Needs: a direct vendor URL and a check of the current service notice.

1. Read the [change notice](https://www.smithery.ai/future).
2. Search the retained directory for an existing correct record.
3. Mark any new-listing or correction request as unresolved until a current submission route is documented.

Confirm: an existing entry uses your live direct URL. Update/claim/metrics: no current route established by the notice. It says hosted connection services, accounts and CLI/SDK services shut down October 31, 2026 while a curated directory remains. Do not use an old CLI publication tutorial. Read November 7 at 18:42:24 UTC.

## 6. Developer catalog cards

Choose these when the audience uses developer tools. A customer's private configuration does not create a public listing.

### GitHub MCP Registry and VS Code

Needs: official Registry record, clear purpose and endpoint/authentication details; repository/license details when applicable.

1. Publish and verify the official Registry entry.
2. Search [GitHub's MCP Registry](https://github.com/mcp).
3. If absent, consult the [GitHub onboarding discussion](https://github.com/github/github-mcp-server/discussions/1257). The accepted collaborator answer describes manual curation before Registry version sync. A community request format is not guaranteed self-service admission.
4. After onboarding, verify the actual listing and subsequent version sync.

Confirm: GitHub entry exists. Update: keep official Registry versions current and verify synchronization. Metrics and onboarding duration: not established. Registry and discussion read November 7 at 18:42:24 and 18:43:27 UTC.

For VS Code, current [MCP documentation](https://code.visualstudio.com/docs/agent-customization/mcp-servers) uses Chat: Open Customizations > Browse > GitHub Feed, with name or @type:mcp search. An individual can use MCP: Add Server for an unlisted remote endpoint. No separate public gallery submission or owner conversion dashboard was established. Read November 7 at 18:42:24 UTC.

### Docker MCP Catalog

Needs: developer, public hosted endpoint, documentation, authentication details and a repository contribution. Remote submissions do not require a Dockerfile.

1. Read the [catalog guide](https://docs.docker.com/ai/mcp-catalog-and-toolkit/catalog/) and [contribution instructions](https://github.com/docker/mcp-registry/blob/main/CONTRIBUTING.md).
2. Prepare the documented server folder and metadata for a remote server, including server.yaml, tools.json containing an empty array ([]), and readme.md linking to documentation.
3. Test only your own authorized configuration. Open a pull request and address checks and review. Keep review credentials out of the public request.
4. Verify the approved entry in Docker's catalog and Toolkit.

Confirm: published catalog entry. Update: use the repository change/review route and check the result. Metrics: no owner enquiry dashboard established. Docker documents availability within twenty-four hours after approval; that is not a review deadline. Remote services do not inherit every security claim made about Docker-built local images. Sources read November 7 at 18:42:24 and 18:43:27 UTC.

### Cursor official marketplace

Needs: a plugin with remote MCP configuration, valid manifest and README in a public Git repository. The distributable marketplace plugin must be open source; this does not establish that every hosted server's implementation must be public.

1. Follow [Cursor plugin documentation](https://cursor.com/docs/plugins) and [reference](https://cursor.com/docs/reference/plugins).
2. Prepare and test the plugin with your own endpoint.
3. Submit its repository at [marketplace publishing](https://cursor.com/marketplace/publish) for manual review.
4. Confirm listing and installation details.

Confirm: official marketplace entry. Update: submit updates through the reviewed plugin route. Metrics/review timetable: not established in the docs read. Cursor's official marketplace and the separate community directory are different services. No community-directory submission recipe is asserted here. Docs read November 7 at 18:43:27 UTC.

### LobeHub

Needs: LobeHub account, GitHub repository ownership/push access and developer preparation under the current publishing instructions.

1. Read [LobeHub publishing documentation](https://market.lobehub.com/s/publish-mcp).
2. Owner completes browser login and GitHub connection. Check for an existing listing and claim it where appropriate.
3. Developer reviews the manifest, version, declared tools and cloudEndpoint for the business's hosted server. An optional generator contacts the server, so authorize it only for your own service.
4. Use the documented plugin publish operation for a new entry, or plugin update for an existing one. These publication actions change the public listing.
5. Verify both the owned-listing view and the public entry.

Confirm: actual listing, not only a command result. Update: push README changes first and update the listing; documentation says README refresh follows within a few minutes while other enrichment runs separately. Metrics: no owner enquiry dashboard established. No paid badge claim is made. Documentation read November 7 at 18:43:27 UTC. Vendor tooling was not installed or run for this kit.

### Microsoft Copilot Studio certification preview

Needs: verified publisher business in Partner Center, Microsoft 365 and Copilot program enrollment, endpoint ownership/control and the current certification materials.

1. Check [current MCP certification requirements](https://learn.microsoft.com/en-us/microsoft-copilot-studio/mcp-certification) before investing in this route.
2. Prepare manifest, tool/authentication configuration including current Azure Key Vault requirements, public docs, support/privacy/terms links and testing evidence. Have the developer check every technical prerequisite.
3. Submit the certification intake form when the owner is ready, then follow Microsoft's instructions if selected. Preview processing capacity is limited.
4. Complete validation and functional/safety review; verify the approved discovery entry on supported Microsoft surfaces.

Confirm: actual approved listing. Update: significant metadata/package changes and added tools require the documented update review. Metrics/review completion time: not established. Read November 7 at 18:44:17 UTC.

Separately, [Copilot Studio can add an existing MCP server to an agent](https://learn.microsoft.com/en-us/microsoft-copilot-studio/agent-extend-action-mcp). That is an integration step, not certification. [Windows on-device discovery](https://learn.microsoft.com/en-us/windows/ai/mcp/overview) registers local or remote servers for a device; the pages checked do not establish another public small-business marketing directory. Both read November 7 at 18:42:24 UTC.

## 7. One listing tracker

Keep one row per public entry and a separate row for each private test connection. Combine extra facts inside a cell rather than adding columns. Do not put private credentials or reviewer passwords here.

| Place and owner | Status and public URL | Version or tool set | Last checked UTC | Next action and due date |
|---|---|---|---|---|
| Official Registry: [role] | [status; URL] | [version] | [time] | [action] |
| Claude: [role] | [status; URL] | [tools] | [time] | [action] |
| ChatGPT personal: [role] | [test status] | [tools] | [time] | [action] |
| ChatGPT public: [role] | [status; URL] | [version] | [time] | [action] |
| Gemini developer: [role] | [test status] | [tools] | [time] | [action] |
| Google preview: [role] | [status; URL] | [version] | [time] | [action] |
| Independent index: [name/role] | [status; URL] | [version] | [time] | [action] |
| Developer catalog: [name/role] | [status; URL] | [version] | [time] | [action] |

Statuses: not started, testing, submitted, live, update pending, paused, not applicable. A submitted form is not live. A listing badge is not a lead.

Same-day sync rule: when a tool changes, update the server, public docs/privacy as needed, Registry version and every listing that repeats its facts on the same day. Record delayed import or review as pending. Assign follow-up until each public entry matches. Preserve compatible server behavior while a platform still serves old approved definitions.

## 8. Copy-ready listing blocks

Original drafting templates, not platform-approved text. Adapt to the destination's current field limits. Do not claim platform endorsement, a security audit or a feature you have not tested.

Short description:

```text
Find [specific public facts].
```

Long description:

```text
[Business name] helps [audience] find [specific facts or options] from
[owned public sources]. Ask about [task] and receive [answer type],
with a link to [business page] for the next step.

[Sign-in requirement]. The connector can [actual supported behavior].
It cannot [specific limits]. Availability is checked at [verified basis]
and does not hold inventory. The person reviews and submits any request
on our website.

Documentation: [public URL]
Privacy: [public URL]
Support: [public URL]
```

Three example prompts:

```text
What [service options] do you offer in [supported area]?
Which [option] fits [ordinary non-sensitive requirement]?
Show me the page to request [service], without submitting anything.
```

Data handling summary:

```text
The server receives [exact necessary inputs]. Sign-in is [requirement].
We use the inputs for [purpose]. Successful and failed requests record
[verified fields or absence of storage], retained for [verified period].
Our hosting and monitoring providers process [actual data categories].
We do not need [unnecessary personal information] for this task.
The website form has its own privacy notice; the person chooses whether
to submit it. Questions go to [public support route].
Last verified: [date, release, responsible role].
```

Reviewer test sheet:

| Scenario | Prompt | Expected tool/result | Human boundary | Result/date |
|---|---|---|---|---|
| Normal request | [prompt] | [tool and facts] | [no submission] | [result] |
| Outside service area | [prompt] | [refusal with next step] | [no record] | [result] |
| Ambiguous choice | [prompt] | [clarification] | [waits for person] | [result] |
| Expired availability | [prompt] | [checked time/refresh] | [no hold] | [result] |
| Unsupported action | [prompt] | [clear limit] | [no message sent] | [result] |

Add destination-specific cases. OpenAI's required five positive and three negative cases are additional structured review materials, not a reason to claim this generic worksheet meets every submission requirement.

## 9. Lead-capture measurement sheet

Define an enquiry as a person-submitted contact, quote or reservation request. Count directory views, tool calls and enquiries separately. None of our listing observations establishes customer leads yet.

Use a non-personal source tag on links returned by tools, such as https://example.com/contact/?source=claude . Where a shared endpoint cannot reliably identify the assistant, use a neutral source such as mcp or a separate documented link route; do not guess. A person can also answer an optional How did you hear about us? form field.

Prefill service category, date or option only when needed. Never put names, email addresses, phone numbers, private messages, tokens or sensitive circumstances in a URL. A prefilled page does not submit or hold anything. The person checks the details and submits. Review links likewise leave writing and posting to the customer.

| Month/source | Returned-link visits | Person-submitted enquiries | Aggregate tool calls | Notes and owner |
|---|---|---|---|---|
| [month]/claude | [count/unknown] | [count] | [count/unknown] | [quality/limitations] |
| [month]/chatgpt | [count/unknown] | [count] | [count/unknown] | [quality/limitations] |
| [month]/mcp | [count/unknown] | [count] | [count/unknown] | [quality/limitations] |
| [month]/self-reported | [not applicable] | [count] | [not applicable] | [deduplication rule] |

Counting rule: [what qualifies and how duplicates are removed]
Consent/privacy rule for website measurement: [actual setup]
Aggregate-counter fields: [tool name, date bucket, success/failure counts]
Raw prompts/arguments stored: [none, or separately justified and disclosed handling]
Directory figures and observation date: [platform, window, known delay]
Enquiry follow-up owner: [role]
Next monthly review: [date]

Do not collect names, emails or phone numbers in tool inputs just to create a lead. OpenAI restricts inputs such as payment-card data, government identifiers, protected health information and authentication secrets including one-time codes; review its [current guidelines](https://developers.openai.com/plugins/plugin-guidelines), read November 7 at 18:43:01 UTC. A useful answer followed by a voluntary website handoff is the starting pattern. Do not send messages on a person's behalf.

## 10. First 30 days audience plan

This is a suggested owner schedule, not a platform review promise. Extend it when review or preview access takes longer.

| When | Owner action | Developer/web action | Done when |
|---|---|---|---|
| Days 1 to 3 | Choose a useful customer question and review limits | Verify server, security and errors | Test record is complete |
| Days 4 to 7 | Approve honest listing text and support route | Publish connect/privacy pages and Registry entry | Pages and active record agree |
| Days 8 to 14 | Check independent entries; prepare Claude submission | Resolve warnings and test account access | Status and follow-up recorded |
| Days 15 to 21 | Test ChatGPT personally; decide public eligibility | Prepare package/review cases; Gemini test | Real execution checked; submissions tracked |
| Days 22 to 30 | Tell existing customers and relevant partners | Check returned links and aggregate counters | First monthly measurement is recorded |

Practical outreach worksheet:

- [ ] Link the connect page from footer and contact page.
- [ ] Link the Registry record and relevant assistant listings from that page.
- [ ] Add a concise connection reference in llms.txt; do not describe this as a ranking guarantee.
- [ ] Where permitted, link from a Google Business Profile or nonprofit profile.
- [ ] Publish a short how-to article or video showing a real customer question and the human handoff.
- [ ] Tell existing customers how to ask about a useful topic such as hours or availability, without implying all assistants can use it automatically.
- [ ] Ask a relevant partner, trade group or association to include the public documentation link.
- [ ] Share the practical example in an existing newsletter or social channel with the owner's approval.
- [ ] Use a chamber or conference talk to demonstrate it when appropriate.
- [ ] If a public repository exists, put setup, scope, privacy and support in its README.
- [ ] Review actual enquiries after a month. Improve confusing answers before buying placement.

## 11. Revenue worksheet

Distinguish a documented paid offering, a plausible income path and an observed result.

Documented paid offerings: [Upfirst](https://upfirst.ai/pricing) sells receptionist plans. [Planning Center](https://www.planningcenter.com/pricing) has free and paid product plans. [Givebutter](https://givebutter.com/pricing) offers optional paid Plus features. These pages were read November 7 at 18:53:25 UTC. No amounts are reproduced. An authenticated connector can serve an existing customer; sign-in does not establish payment or a new sale caused by the listing.

[Claude Marketplace](https://claude.com/resources/articles/claude-marketplace) and its [November update](https://claude.com/resources/articles/new-agents-and-products-claude-marketplace), read at 18:42:31 and 18:42:33 UTC, describe a route where a vendor applies to list a Claude-powered product, and eligible enterprise customers can put part of their committed Anthropic spend toward it. Ordinary connector publication does not guarantee that route or any sale.

Plausible model to test: useful answer, own-page link, person-submitted enquiry, accepted quote or booking, then payment. Our [TF Rooms example](https://tfrooms.com/connect.html), documentation read at 18:42:58 UTC, takes reservations with staff confirmation and payment at the property, without collecting payment through the connector. A free checker could produce consulting enquiries; a nonprofit's own donation or event page could receive gifts or ticket sales. These are possible paths, not measured income claims.

Apply [OpenAI commerce and checkout rules](https://developers.openai.com/plugins/plugin-guidelines), read at 18:43:01 UTC, before adapting any model. Physical-goods eligibility and own-domain checkout requirements are separate from whether a connector technically works. Selected-partner Instant Checkout is not general eligibility. A plugin must provide standalone value rather than exist primarily as advertising.

| Month/source tag | Completed sales or gifts | Received revenue less refunds | Maintenance cost | Evidence and limitations |
|---|---|---|---|---|
| [month]/claude | [count] | [actual amount/currency] | [cost] | [private reconciliation] |
| [month]/chatgpt | [count] | [actual amount/currency] | [cost] | [private reconciliation] |
| [month]/unknown | [count] | [actual amount/currency] | [cost] | [attribution gap] |

- Documented offering and source: [public page, read date, what it proves]
- Proposed earning path: [task, destination, paid service or contribution]
- Received-money definition: [payment received, refunds, reporting period]
- Source assignment: [known tag or voluntary answer; unknown when unproved]
- Reconciliation owner: [role]
- Current evidence: [measured result, or no evidence yet]

Keep customer identities, payment data and private transaction records in the normal protected business system. Share aggregate totals only. Count clicks, enquiries, quotes and pending bookings separately from received money.

No evidence yet: we found no connector-attributed revenue figures for the businesses named here, and our own directory observations do not yet establish listing-attributed income. A listing without a useful next step supplies no demonstrated earning path.

## Sources and read dates

All dates are November 7, 2026, UTC. Requirements may change. The linked official pages in each card are the source for its claims, not a guarantee of eligibility or acceptance.

- Official Registry: [quickstart](https://modelcontextprotocol.io/registry/quickstart) 18:42:58; [authentication](https://modelcontextprotocol.io/registry/authentication) 18:42:59; [remote servers](https://modelcontextprotocol.io/registry/remote-servers) 18:43:01; [versioning](https://modelcontextprotocol.io/registry/versioning) 18:43:04; [scope](https://modelcontextprotocol.io/registry/about) 18:44:30.
- Claude: [directory overview](https://claude.com/docs/directory) 18:42:27; [submission](https://claude.com/docs/connectors/building/submission.md) 18:42:59; [verification](https://claude.com/docs/connectors/verification.md) 18:43:00; [metrics](https://claude.com/docs/connectors/building/managing-your-listing.md) 18:43:01; [updates](https://claude.com/docs/connectors/building/after-publishing.md) 18:43:03.
- OpenAI: [submission](https://developers.openai.com/plugins/deploy/submission) 18:42:58; [validation](https://developers.openai.com/plugins/deploy/submission-errors) 18:42:59; [guidelines](https://developers.openai.com/plugins/plugin-guidelines) 18:43:01; [connection/testing](https://developers.openai.com/plugins/deploy/connect-chatgpt) 18:43:03.
- Google: [preview](https://developers.google.com/connected-apps) 18:42:34; [developer testing](https://developers.google.com/connected-apps/mcp/dev-mode/usage) 18:43:04; [publication](https://developers.google.com/connected-apps/gpc/perform-the-steps) 18:43:05; [prerequisites](https://developers.google.com/connected-apps/gpc/prerequisites) 18:43:07.
- Independent services: Glama, mcptry, MCP Market and Smithery sources in their cards read 18:42:24; Glama/mcptry methodology read 18:43:27. PulseMCP browser read 18:44:05; mcp.so browser read 18:44:12.
- Developer catalogs: GitHub Registry, VS Code, Docker catalog, Copilot Studio integration and Windows overview read 18:42:24; GitHub onboarding, Docker contribution instructions, Cursor docs and LobeHub publishing read 18:43:27; Microsoft certification read 18:44:17.
- MCP-Sentinel [license API](https://api.github.com/repos/BashaarJavaid/MCP-Sentinel/license): MIT, 18:43:04. No third-party project was installed or run.
- Revenue: Upfirst, Planning Center and Givebutter pricing pages linked in section 11, read 18:53:25; paid offerings established, amounts and connector-attributed revenue not claimed.

Validation scope: shell example syntax only; no placeholder server requests, publication commands, payments, form submissions or third-party MCP calls were performed. Copy blocks are original prose templates and contain no executable configuration. Owner-specific answers, privacy claims and listing outcomes still require verification.