TYPO3 MCP clients compared: which AI assistant works with your TYPO3?

Your organisation already pays for an AI assistant, and the question is whether it can work in your TYPO3.

This comparison of TYPO3 MCP clients answers it for eight assistants: Claude, Microsoft Copilot, Cursor, VS Code with GitHub Copilot, Claude Code, ChatGPT, Gemini and PhpStorm. MCP, the Model Context Protocol, is the open standard an assistant uses to call the tools of another system, and AI Foundation, T3Planet's free and open source extension, adds an MCP server to TYPO3. For each assistant you see the plan its vendor names, how it signs in, what your admin controls, and its status with AI Foundation.

Two terms first, because they sound alike. Your AI assistant is the app you chat or code in, such as Claude or Cursor. AI Assistant, with capitals, is our licensed TYPO3 extension for writing, SEO and translation, and nothing in this post needs it.

An AI assistant works with TYPO3 through AI Foundation's MCP server: it connects over Streamable HTTP and signs in with OAuth, which means you approve it in TYPO3 and it receives a token instead of your password. AI Foundation ships setup guides for Claude, Microsoft Copilot through Copilot Studio, Cursor and VS Code with GitHub Copilot. ChatGPT, Claude Code, Gemini and PhpStorm passed our demo test on 30 September 2026. PhpStorm is the one exception to the OAuth sign-in: JetBrains documents none for MCP servers yet, so it connects through a token URL.

AI Foundation is free and open source, no licence key, no registration, and the MCP server is part of it. How MCP works in general is in our explainer on TYPO3 MCP. This post stays with one question: which client fits your team.

Each row carries one of two statuses. Setup guide in AI Foundation means the product ships setup steps for the assistant, in the Connect tab of its MCP server or in its documentation. Verified on our demo means our team connected the assistant to our demo TYPO3 and it passed the test described further down: sign in, read a page, change one header on the live demo site, check it on the page and put it back. Every row links our guide for that assistant. Rows are sorted by status, then by name.

The table ranks nothing. It does not say which assistant writes the better text, because that depends on the model behind it and on your prompts. Each assistant's plan is paid to its vendor; AI Foundation costs nothing.

Editors and marketers usually meet their assistant in the browser or in Microsoft 365, and the contract is already signed. What matters is who sets it up and who approves it.

• Claude connects through a custom connector. On Team and Enterprise, an Owner adds your TYPO3 MCP server once for the organisation, and each person then connects and approves the sign-in in TYPO3. On Pro and Max, and with one connector on Free, you add it yourself under Customize > Connectors. Our guide Connect Claude to TYPO3 covers both.
• Microsoft Copilot reaches TYPO3 through an agent built in Copilot Studio. A maker adds the MCP server as a tool, with OAuth set to dynamic discovery, and your team uses the agent where it is published. Your Power Platform data policies apply to it as well. AI Foundation's setup guide for it is called Copilot Studio; ours is Connect Microsoft Copilot to TYPO3.
• ChatGPT takes MCP servers as apps in developer mode, on the web. OpenAI's pages disagree on the plans: its help centre names Business, Enterprise and Edu for full MCP including writes, its developer docs add Plus and Pro. ChatGPT passed our demo test on 30 September 2026, and our guide Connect ChatGPT to TYPO3 shows the steps. Check what your own plan includes on OpenAI's page before you start.

AI Foundation's Connect tab carries setup guides for many more clients, from Claude Desktop and Copilot Studio to n8n and the MCP Inspector, so check it for your assistant before you look anywhere else.

In an editor or a terminal, the MCP server is one entry in a configuration file, and the sign-in opens in your browser.

• Cursor and VS Code with GitHub Copilot take a remote MCP server with OAuth: you give them the server URL of your TYPO3, and they handle the sign-in themselves. On GitHub Copilot Business and Enterprise, ask your GitHub admin to switch on the policy "MCP servers in Copilot" first; it is off by default. Our guides: Connect Cursor to TYPO3 and Connect VS Code to TYPO3.
• Claude Code adds a remote server with claude mcp add --transport http, and the /mcp command starts the sign-in in your browser. AI Foundation's Connect tab has no preset for it, and Claude Code passed our demo test on 30 September 2026; our guide Connect Claude Code to TYPO3 has the steps. If you want Claude Code to build TYPO3 extensions rather than work on a site, our Claude Code & TYPO3 guide covers that.
• PhpStorm with JetBrains AI Assistant takes remote MCP servers, but JetBrains documents no OAuth sign-in for them yet. For clients like that, AI Foundation offers a second route: a token URL that the mcp-remote package bridges to your IDE. A token URL works like a password, so keep it out of Git, screenshots and chats. PhpStorm passed our demo test on 30 September 2026, and our guide Connect PhpStorm to TYPO3 shows the token route.

Like any MCP client comparison, this one ages: plans and features move every month. These five questions do not, and they also work for an assistant this table does not list:

1. Does our plan include remote MCP servers? Read the vendor's own page and note the date; plan pages change often.
2. Does it sign in with OAuth, or does it need a token in a URL? With OAuth you approve it in TYPO3 and revoke it there. A token URL has to be stored like a password.
3. Does it ask before it changes something? Look for a confirmation step for write actions, and for who can switch it off.
4. Can our admin allow, limit or switch it off? Look for organisation settings and allowlists on the vendor's side. On yours, AI Foundation lists every client's token and lets you revoke it.
5. Where does what it reads go? In context mode, what your assistant reads through MCP goes to its vendor, as with any chat. In native mode, TYPO3 generates text with the AI provider set in AI Foundation, with your own API keys by default; choosing that provider is the subject of Multi-LLM for TYPO3.

An assistant without a setup guide in AI Foundation goes through the same test on our demo TYPO3 before its guide goes live. On 30 September 2026, ChatGPT, Claude Code, Gemini and PhpStorm passed it:

1. Add the demo's MCP server to the assistant and approve the sign-in in TYPO3.
2. Ask it to read a page and list its content elements.
3. Ask it to change the header of one content element on the live demo site, check the change on the page, and have it put the old header back.
4. Keep the session log: prompts, tool calls and answers.

The test runs as one administrator on the live demo site, and every change is put back before the next assistant. Gemini needs a note of its own, because the name covers three products. The Gemini app takes custom MCP apps only for personal Google Accounts in the US. Gemini Enterprise is set up by an administrator and signs in with a client ID and a client secret. Antigravity is Google's developer IDE and command line. Check Google's requirements for the one your team uses before you start; our guide Connect Gemini to TYPO3 names them.

The server side is the same for all TYPO3 MCP clients, which is why the choice of client matters less than it seems:

• It acts as a TYPO3 backend user. The assistant works with the rights of the backend user who approved its sign-in, through the tools AI Foundation provides. AI Foundation and the AI Universe extensions together bring 100+ MCP tools.
• One token per client. MCP Server > OAuth Tokens lists every connected client with its backend user, when its token was created and last used, and when it expires. Revoke one, and the others keep working.
• One set of rules for all of them. PKCE, default scopes, an IP allowlist and mutual TLS are set once in MCP Server > Security. Our security episode goes through each setting.
• Check every change, and know how to put it back. Before a change, ask the assistant to show the current value; after it, check the page, and ask it to restore the old value if needed. When it writes with records_apply, ask for a dry run first and keep the batch ID for records_undo, which takes the whole batch back, as episode 3 shows. Where safety matters, AI Foundation can hold changes to pages and content in a TYPO3 draft workspace for review, which our workspaces episode shows.

This one-minute video draws the figures of this post: one MCP server for your TYPO3, two routes into it (OAuth sign-in for seven assistants, a token URL for PhpStorm), and the Connect and OAuth Tokens tabs of AI Foundation. Open the video page with transcript and chapters.

To connect an AI assistant to TYPO3, you need one thing on the TYPO3 side: an MCP server. AI Foundation adds it with one Composer command. TYPO3 MCP setup, the second post of this episode, shows how to install AI Foundation and switch on the server; the guide for your assistant takes it from there, and episode 3 brings your first everyday edits from chat. Every guide and episode is on the hub of MCP in Practice for TYPO3.

AI Foundation's core tools are free. AI Assistant, AI Accessibility, AI Chatbot and AI Search add tools of their own and need their licences.

The series comes from T3Planet by NITSAN Technologies, a TYPO3 Association Gold Member with more than 15 years of TYPO3 work, and its episodes are written by Jürgen, our AI Consultant for TYPO3. Every status in this post is dated and comes from our demo install, from the setup guides AI Foundation ships, or from the vendor's own page.

Get AI Foundation for TYPO3