Which Claude and ChatGPT Plans Support MCP Connectors in 2026 (Free Plans Included)

Ask the web whether your Claude or ChatGPT plan can use MCP connectors and you get answers that were true at some point between early 2025 and now, presented as current. The features keep getting renamed (ChatGPT's connectors became "apps" in December 2025), the plan gates have moved more than once, and some of the wrong answers sit on official-looking pages with fresh "last updated" stamps.

I connect agents to a remote MCP server every day, so this week I went through the official documentation for every plan tier. Every claim below links to a primary source, and where the primary sources disagree, I say so and date the claims.

The matrix, as of June 2026

The Claude rows come from Anthropic's custom connectors article (updated April 2, 2026) and the connectors overview. The ChatGPT rows come from the per-plan capability table in Apps in ChatGPT (updated the day I checked) and the two developer mode documents discussed below.

Directory connectors are the curated, one-click kind. Custom remote MCP is the column that matters if you want to connect your own server, and it is where the documentation gets messy.

Claude: the free plan gets one custom connector

This is the fact most third-party guides have wrong. Anthropic's support article states it plainly: "Custom connectors using remote MCP are available on Claude, Cowork, and Claude Desktop for users on Free, Pro, Max, Team, and Enterprise plans," with one cap: "Free users are limited to one custom connector." The feature is labeled beta, and the article was updated April 2, 2026.

On Free, Pro, and Max you add the connector yourself under Settings > Connectors. On Team and Enterprise, only Owners can add custom connectors to the organization; members then connect and enable them individually, and Owners can restrict which actions a connector may take, for example allowing reads while blocking writes org-wide. One oddity: the setup steps in the article cover Pro/Max and Team/Enterprise, but no Free-specific walkthrough exists yet. In practice the Pro/Max steps are the ones to follow.

Claude: where the connector actually runs

A detail that trips people up: custom connectors are brokered through your Claude account, and the connection to your MCP server originates from Anthropic's cloud, not from your machine. This holds on every client, including Claude Desktop and Cowork. Your server must be reachable from the public internet; a server behind a VPN or corporate firewall will not connect even if your own laptop can reach it.

Local MCP servers are a separate mechanism: configured in claude_desktop_config.json, they run on your machine, work only in Claude Desktop, and the docs state no plan restriction for them. Claude Code supports both local and remote MCP servers with no MCP-specific gate, but since Claude Code itself comes with Pro and Max plans, premium seats, or API billing, it is not a free-plan path.

The official-looking page that is wrong

This is why you cannot settle the question with one search. Google's documentation for the Drive remote MCP server states: "To use the Google Drive remote MCP server with Claude.ai or Claude Desktop, you must have the Claude Enterprise, Pro, Max, or Team plan." That page carried a "last updated" stamp of June 5, 2026 when I checked, a week before this article. It is still wrong: Anthropic opened custom connectors to free users in April 2026, and Google's page has not caught up.

A freshly stamped page on developers.google.com outranks most things, and language models grounded on it will repeat the stale claim. When a platform's own support article and a partner's integration docs disagree, the platform wins.

ChatGPT: the free plan gets connectors, not custom ones

OpenAI's plan table in Apps in ChatGPT is unambiguous: Free and Go plans get directory apps, including interactive apps and write actions, with limited search and deep research, and no Custom (MCP) capability at all. The app directory is the mainstream path and requires no developer mode: browse, click Connect, authorize. Per the same article, "Apps are available to all logged-in ChatGPT users, with some exceptions," the exceptions being regional gaps in the EEA, GB, and Switzerland and some features reserved for paid plans.

So on a free ChatGPT account you can use connectors someone else published to the directory, but there is no supported way to point ChatGPT at your own MCP server. Custom MCP starts at Plus.

ChatGPT: developer mode, where the docs disagree with each other

Custom MCP servers in ChatGPT go through developer mode, and OpenAI's two documents about it do not match. The platform docs say developer mode is "Available to Pro, Plus, Business, Enterprise, and Education accounts on the web," enabled under Settings > Apps > Advanced settings, with "write actions ... available, subject to confirmation settings." The help center article, updated days before I checked, says full MCP including write actions is "rolling out in beta to ChatGPT Business, Enterprise, and Edu plans" and that "Pro users can connect MCPs with read/fetch permissions in developer mode." Plus appears in the capability table but not in that FAQ at all.

My read: Plus and Pro get developer mode with custom connectors, write support is in flux, and the help center is the more recently maintained source. The gating is rough in practice too: an official forum thread from April 2026 documents a Pro account with no developer mode toggle anywhere, staff confirming it should be there, and no resolution posted.

ChatGPT: the fine print once you are in

Three more constraints from the developer mode article, each easy to miss. Custom MCP apps are web only; the FAQ on mobile availability reads "No - web only." Agent mode will not use custom apps, and deep research will use them for read and fetch actions only, never writes. And ChatGPT cannot talk to a local MCP server directly; it connects to remote servers, with a Secure MCP Tunnel offered for servers on private networks.

On workspace plans the human gating is strict as well. On Business, only admins and owners can use developer mode, each one enabling it for themselves. On Enterprise and Edu, admins can extend developer access to chosen members through RBAC. A regular member of a Business workspace has no self-serve path to a custom connector.

What a remote MCP connector is, in one paragraph

MCP (Model Context Protocol) is an open standard that lets an AI assistant call tools that live outside it. A remote MCP connector is simply a URL: a server somewhere on the internet that tells the assistant "here are the actions I offer," such as searching a database, creating a ticket, or saving a file. When you add a custom connector, you paste in that URL, usually sign in once via OAuth, and from then on the assistant can use those actions inside a normal conversation. No code runs on your machine, which is why the plan gates above are the whole story: the only question is whether your subscription lets you paste the URL.

How to re-check this when it changes

This will drift. Both vendors shipped plan-gate changes within the last few months, and both label the current behavior beta. When you need a fresh answer, skip the search results and read three pages: Anthropic's custom connectors article, OpenAI's Apps in ChatGPT plan table, and OpenAI's developer mode article. Trust the dated support article over any third-party page, including a big vendor's.

The fastest empirical check is to add a connector and watch what happens. Caipi is my product and happens to be a convenient test subject: a remote MCP connector that gives your agent a hosted workspace to write files into, which also means the work survives outside the chat window. If your plan can add it, your plan supports custom MCP. If not, the matrix above tells you which upgrade changes that.

Everything here was checked against the official documentation linked above in June 2026.