Owner’s Desk
Our own OAuth 2.1 connector. It reaches the owner’s Stripe and Google accounts, every write waits for a confirm token, and every action lands in an audit log.
Connector builds / Claude connector vs ChatGPT app
Claude connector or ChatGPT app? Build one server.
A software company does not have to choose between a Claude connector and a ChatGPT app. Both are a remote MCP server; the same server, built once, can be submitted to Anthropic’s Claude directory and to OpenAI’s directory shared by ChatGPT and Codex. What differs is the review. OpenAI asks for domain verification, terms of service, a test account without MFA, five positive and three negative test cases and a video. Anthropic asks for a fully populated test account and seven compliance acknowledgments, and lists you as Community first.
Nine of our tools are listed in Anthropic’s Claude directory.
At a glance
- Same
- MCP server, OAuth, read and write tools, annotations
- Claude adds
- Seven compliance acknowledgments, Community then Verified
- ChatGPT adds
- Domain token, terms URL, test cases, a video
- My advice
- One build, two submissions
Checked October 12, 2026
Side by side
The two submissions, side by side
| Factor | Anthropic’s Claude directory | ChatGPT (OpenAI’s plugin directory) |
|---|---|---|
| What you submit | A remote MCP server, as an MCP connector, in the developer portal | A plugin, which can carry one connected MCP server, skills and optional UI |
| Who can submit | Anyone on a paid Claude plan | Organization owners, or members with Apps Management Write |
| Sign-in | OAuth 2.0 for authenticated services, or none for public data | OAuth; the portal has no API-key or bearer-token input |
| Tool hints | A title plus readOnlyHint or destructiveHint on every tool | readOnlyHint, destructiveHint and openWorldHint, set accurately |
| Proof you own it | The server domain should match your service, calling your first-party API | A challenge token hosted on your domain, and verification to publish under your company name |
| Public URLs | Docs, privacy policy, support contact | Website, support, privacy policy and terms of service |
| Reviewer access | Credentials for a fully populated test account | A test account that works immediately, with no MFA |
| Testing evidence | Confirm you ran every tool in MCP Inspector or as a custom connector | Five positive and three negative test cases, plus a video walkthrough |
| After review | Listed as Community by default; some escalate to Verified | Feedback by email; once approved, you choose when to publish |
Which to file first
File both, but start where your users are.
Ask your support and success teams which assistant your customers mention by name, then file that submission first and the other one the same month. The server does not change between them.
The one design choice that matters for both is the approval step. Claude prompts the customer on any tool marked destructive, and OpenAI reviewers check that your hints match what each tool really does. A write that drafts first and runs on a confirm token satisfies both, and it is how Owner’s Desk handles every action.
What never changes
The parts both reviewers read.
Whichever directory you file with, these are read closely:
- Tool names and descriptions that say exactly what the tool does, with no instructions aimed at the model
- Writes split from reads, never one catch-all request tool
- Errors that explain themselves
- A privacy policy that names every category of personal data the connector touches
- Listed in Anthropic’s Claude directory
- 9
- In review with Anthropic
- 4
- In the official MCP Registry
- 15
- Tools Claude can call
- 69
What we have shipped
Built and running, not a deck.
AI Front Door
One multi-tenant deploy that gives every business its own connector, white label, under an agency’s brand.
Small Business Agent Bench
Our open benchmark of how well AI agents handle real small-business work: 200 tasks, sandboxed tools, deterministic grading.
The package that fits
Connector Build
Connector Build includes both submissions. A second product or system is Connector Suite.
Your customers do their real work in your product from inside the assistant they already use. Set scope, set price, quoted in the conversation, and changes to what we built are included.
Straight answers
What teams ask next.
Do we need two codebases for Claude and ChatGPT?
No. One MCP server serves both. The differences live in the submission: listing URLs, test cases, verification and labels.
Which directory is harder to get into?
They are hard in different places. OpenAI front-loads evidence: test cases, a video, domain verification. Anthropic lists a passing server as Community by default after an automatic scan, and saves its deepest testing for Verified review.
Can a connector move money in either one?
Anthropic’s review criteria do not accept connectors that transfer money or other financial assets, and its compliance step asks about financial transactions. I design payment jobs as reads and drafts, never as transfers.
Sources
I wrote this page for Modern Mustard Seed, and we sell the service it describes. Every fact about Anthropic, OpenAI or another company was read from the page linked here on October 12, 2026.
- Anthropic: Submit a connector to the directory (opens in a new tab)
- Anthropic: Connector pre-submission checklist (opens in a new tab)
- Anthropic: Connector verification (Community and Verified labels) (opens in a new tab)
- OpenAI: Upload and submit your plugin (opens in a new tab)
- OpenAI: Remote MCP server review requirements (opens in a new tab)
One next step
Send me your API docs.
Tell me your product and where your API reference lives. I will come back with the jobs your customers would do in Claude and ChatGPT, the tools I would build, and the package that fits. Managed for you, end to end, and the code is yours.
Claude is a product of Anthropic, and ChatGPT is a product of OpenAI. Modern Mustard Seed is an independent studio and is not affiliated with or endorsed by either.