Sound familiar?
- Your customers already work in ChatGPT or Claude and ask why your product is not there
- Staff copy data out of your systems into an AI chat by hand, with no control over what leaves
- You built an MCP server but it was rejected from a directory and you are not sure why
- You want an interactive card or form inside the chat, not just text answers
- Marketing wants a listing in ChatGPT; IT wants to know exactly what data it returns
- Nobody on the team has followed the platform rules for privacy policies, test accounts and annotations before
Key facts
- Both platforms build on the Model Context Protocol, so one well-designed server can serve both
- OpenAI opened app submissions for ChatGPT in December 2025 and, from July 2026, lists them in a Plugin directory
- OpenAI's review checks the readOnlyHint, destructiveHint and openWorldHint annotation on every tool
- OpenAI's submission docs say projects with EU data residency cannot submit MCP plugins for now
- Anthropic says Claude custom connectors work on Free (one connector), Pro, Max, Team and Enterprise plans
- Anthropic's Connectors Directory lists approved servers as Community connectors, with a further Verified review for some
- Internal-only connectors for your own staff do not need a public directory listing
What these integrations are, and what they are called now
Both OpenAI and Anthropic let outside developers plug software into their assistants, and both use the Model Context Protocol underneath. The names keep changing, so here is where they stood when we checked on 11 November 2026.
- ChatGPT: OpenAI launched the Apps SDK in preview in 2025 and opened app submissions in December 2025. OpenAI's developer site now frames this work as plugins, which it describes as packages of skills, MCP servers and optional UI, listed in a directory shared by ChatGPT and Codex. Reporting on OpenAI's help centre says the app directory moved to the Plugin directory on 9 July 2026. An app built with the Apps SDK sits inside a plugin.
- Claude: Anthropic calls them connectors. A custom connector is any remote MCP server a user or admin adds by URL. The Connectors Directory is the public listing.
Because the terms shift, we confirm the current product names, plan limits and review rules at the start of every project rather than relying on last quarter's blog posts.
Private connector or public listing?
Most businesses we speak to want one of two things.
Internal use. Your staff use Claude or ChatGPT at work and you want them to reach your CRM, stock system or document store under your rules. This does not need a directory listing. On Claude Team and Enterprise, Anthropic's help centre says an Owner adds the custom connector once for the organisation and members then connect individually with their own sign-in. Plan rules for custom apps in ChatGPT differ by plan and change often, so we check OpenAI's current help pages during scoping.
Public listing. You sell software and want any ChatGPT or Claude user to find and connect it. That means meeting each platform's directory rules, passing review and maintaining the listing. It is worth it if your customers already live in those tools.
The engineering is mostly shared: one MCP server, one permission model, one audit log. Only the packaging and review differ.
What the review processes ask for
Both platforms publish their requirements, and most rejections come from the same handful of gaps.
- Accurate tool annotations. OpenAI's review requirements say
readOnlyHintis true only if a tool strictly reads data,destructiveHintis true if it can cause irreversible outcomes, andopenWorldHintis true if it reaches the public internet. Anthropic's directory docs also ask for a title and read-only or destructive annotation on every tool. - A working reviewer account. OpenAI asks for a demo login that works without extra setup or MFA codes; Anthropic asks for test credentials for a fully populated account.
- Minimal data. OpenAI's plugin guidelines say to collect only what the function needs and never to collect payment card data, health information, government IDs or passwords.
- Verified publisher. OpenAI rejects publisher names that have not completed individual or business verification on its platform.
- Commerce limits. OpenAI's guidelines currently limit commerce to physical goods, with checkout on your own domain. Digital subscriptions and credits are not allowed.
One UK-relevant catch: OpenAI's submission docs say projects using EU data residency cannot submit MCP plugins for now. If your OpenAI organisation is set up that way, we plan around it.
Interfaces inside the chat
Text answers are fine for lookups, but some jobs need a table, a map, a booking form or a product card. OpenAI's Apps SDK lets an app return UI components that render in ChatGPT. On the open standard side, the MCP project now documents an optional MCP Apps extension for interactive elements such as charts and forms rendered inline in a conversation. Client support for these interface features varies, so we always ship a text fallback that works everywhere and add rich UI where the client supports it.
We build these components in React and TypeScript, keep them small and fast, and make sure they still make sense when the model only reads the structured data behind them.
Why us
Most of a good ChatGPT app or Claude connector is ordinary integration engineering: clean APIs, careful OAuth, sensible permissions, logging and support. We have been building integrations since the KashFlow and OpenCart days and still do that work daily. We also build on these AI platforms ourselves: dijitul.ai is our UK-hosted Claude API gateway, so we follow the platform changes closely because they affect our own service.
Every project starts with a free chat, then a short scoping stage that settles which platforms, which tools and whether to list publicly, then a fixed-price quote.
What we deliver
- A decision on public listing versus private, internal connector for each platform
- A remote MCP server with OAuth sign-in and accurately annotated tools
- Optional interactive UI components for ChatGPT and for MCP Apps-capable clients
- A privacy policy section describing exactly which personal data the tools return
- A populated reviewer test account without multi-factor prompts, as both platforms ask
- Directory submissions for OpenAI and Anthropic, and fixes for any review feedback
- Usage logging and an update process for when the platforms change
How it works and what it costs
Every project gets a fixed-price quote after a free initial chat and a short scoping stage. You own the code and the data.
Free chat
Tell us the problem in plain English: what you do now, what goes wrong and what "better" looks like. No charge, no obligation.
Scoping
We map the processes, systems and data involved, agree what is in and out, and write it down so there are no surprises.
Fixed-price quote
You get a fixed price for the agreed scope, or a phased plan for bigger builds, so you can start small and prove it works.
Build and test
We build in short stages you can see and try, test against real data, then go live carefully with a rollback plan.
Hand over and look after
You own the code and the data. We can host it, support it and keep improving it, or hand it to your own team.
Frequently asked questions
What is the difference between a ChatGPT app and a Claude connector?
Both connect outside software to an AI assistant, and both are built on the Model Context Protocol. OpenAI uses the Apps SDK and now lists them in its Plugin directory; Anthropic calls them connectors and lists them in its Connectors Directory. dijitul developments usually builds one MCP server and packages it for each platform.
Do we need to be listed in a directory?
No. If only your own staff will use it, a private custom connector is enough. Anthropic lets Team and Enterprise Owners add a remote MCP server for their organisation. A public listing is worth it when you sell software and want any ChatGPT or Claude user to find and connect it.
Can a ChatGPT app take payments?
Within limits. OpenAI's current plugin guidelines restrict commerce to physical goods, with checkout on your own domain, and do not allow digital subscriptions, credits or tokens. If selling through AI assistants is the goal, see our agentic commerce integration page for the protocols involved.
Why do directory reviews get rejected?
Common reasons are inaccurate read-only or destructive annotations on tools, a test account that needs MFA or has no data, tools that return generic errors, missing privacy policy detail and unverified publisher names. dijitul developments checks against each platform's published requirements before submitting.
Is this available to UK businesses?
Yes, both platforms are used in the UK. Check one detail: OpenAI's submission docs say projects with EU data residency cannot submit MCP plugins for now. Directory submissions also ask which countries you want the listing available in. We confirm current rules at scoping.
How is a ChatGPT or Claude integration priced?
dijitul developments gives a fixed-price quote after a free chat and a scoping stage that settles the platforms, the tools and whether you need a public listing. Directory review time is outside our control, so we plan the timeline around it rather than promising a launch date.
Related
Tell us what you need to build
Free chat, clear scope, fixed-price quote. You own everything we build.