Which tools can connect today?

Any client that speaks the Model Context Protocol. Claude has a published connector; ChatGPT and Codex connect through their own MCP flow; other tools and IDEs connect to the remote server directly. The installation page shows one route per ecosystem rather than a generic list.

Do I have to install a plugin?

No. The plugin is optional and only offered where the client confirms it is supported. The connection alone is what makes the memory readable.

Is context copied between tools?

Nothing is copied. There is one project on the server, and each assistant reads it — so there is no sync to fall behind.

Answers

Shared context across Claude, ChatGPT and Codex

One connection, read by every assistant. Arroway runs as a remote MCP server, so Claude, ChatGPT, Codex and any other client that speaks MCP read and write the same project memory — the context does not live inside one vendor, and moving tools does not restart it.

The server is the universal part; the plugin is a reinforcement

The remote server is what every ecosystem gets. A plugin exists only for the surfaces where the client confirms support, and it carries the trigger and the transport — never the judgement. Everything that could need to change later lives on the server, so one release reaches everyone instead of waiting for each person to update something on their machine.

Each person authorizes with their own account

A team or enterprise workspace can make the connection available, and every person still signs in with their own Arroway account. That is what keeps authorship individual: the record says which person wrote a line, not which workspace did.

The same read, wherever it is asked from

The read is assembled on the server and the same material is served to any client, adapted to what that client can transport — a long read is delivered in parts with a cursor, so nothing is silently dropped or repeated. Refusing a request whole, with the number of units left out, is preferred over answering success for a part.

What it looks like in practice

You settle an approach in Claude in the morning. In the afternoon you open Codex on the same repository, it reads the project, and the approach is already there — including the part you decided not to do, so it is not proposed to you again.

Questions people ask about this

Which tools can connect today?: Any client that speaks the Model Context Protocol. Claude has a published connector; ChatGPT and Codex connect through their own MCP flow; other tools and IDEs connect to the remote server directly. The installation page shows one route per ecosystem rather than a generic list.
Do I have to install a plugin?: No. The plugin is optional and only offered where the client confirms it is supported. The connection alone is what makes the memory readable.
Is context copied between tools?: Nothing is copied. There is one project on the server, and each assistant reads it — so there is no sync to fall behind.

Install Arroway See how it works →

Where this is verifiable

The distribution decision recorded in the product documentation (docs/produto/distribuicao.md in the repository), the published Claude connector, and the installation page on this site. Support per surface is stated only where the client confirms it.

https://www.arroway.app/en/answers/shared-context-across-ai-tools