/reright

Connect from the cloud

Claude.ai, Claude Desktop, ChatGPT and Claude Projects run on the provider's servers, not on your machine. The hooks the installer sets up cannot reach them. You can still add reright as a connector, so the assistant can send its text for review. This page shows how, and says plainly what that does and does not do.

What a connector can and cannot do

A connector gives the assistant two tools, submit_for_review and wait_for_review. With the instructions below, it asks the assistant to submit text first and to use only the approved version.

It does not block anything. A chat client has no hook that can stop a message, a commit or an email. If the assistant ignores the instructions, or you tell it to skip the review, the text goes out and reright never sees it. Use the connector for text you want reviewed, not for text that must never go out unreviewed. Blocking only works where a hook can run, which means the agents the installer sets up on your own machine.

The server URL

Every client asks for the same address. Sign-in is OAuth with your reright account, so there is no token to copy.

https://app.reright.it/mcp

If the client asks how it should identify itself to the server, choose to register automatically (dynamic client registration).

Claude.ai and Claude Desktop

The steps are the same in the browser and in the desktop app.

  1. Open Customize, then Connectors.
  2. Select Add custom connector.
  3. Paste the server URL above and select Add.
  4. Select Connect and sign in to reright when the page opens.
  5. In a chat, open the + menu, choose Connectors and switch reright on for that conversation.

A Free plan can add one custom connector. On Team and Enterprise plans an Owner adds it once under Organization settings, Connectors, Add, Custom, Web, and each member then selects Connect with their own account. Menu names can change, so Anthropic's guide is the reference.

Claude can ask before each tool call. Leave that on for the first few reviews, and only choose "Always allow" once you trust how it behaves.

ChatGPT

ChatGPT adds remote servers in developer mode, which is limited to some plans and can be turned off by a workspace admin.

  1. Turn on developer mode in ChatGPT settings, under Connectors.
  2. Create a new connector, give it a name such as reright, and paste the server URL above.
  3. Choose OAuth, create it, and sign in to reright.
  4. In a new chat, switch the connector on for that chat.

OpenAI's developer mode guide has the current steps.

Claude Code only

If you use Claude Code, run the installer instead. It adds the same server and the hooks that block unapproved text. To add only the tools, without hooks:

claude mcp add --transport http reright https://app.reright.it/mcp

Projects and custom instructions

A connector only makes the tools available. The assistant has to be told to use them. Add this to the instructions of a Claude Project, to ChatGPT's custom instructions or project instructions, or to the system prompt of any other assistant:

Before you send, post or commit any text a person will read (emails, messages, comments, PR or commit text), call submit_for_review on the reright server with the kind, the exact text, where it goes, and a short context. Give me the review link once, then call wait_for_review with the id until the status is no longer pending. If it is approved, use final_text exactly, character for character. If it is rejected, stop, do not retry or rephrase, and wait for my instructions.

This is a short version for chat clients. The full rules text, which the installer writes for coding agents, is in the docs.

Check that it works

  1. In a new chat with the connector on, write: "Use reright to review this message before you show it to me: Hello, the report is ready."
  2. The assistant should reply with a review link. Open it, edit the text if you like, and approve.
  3. The assistant should then show you the approved text, unchanged.

If no link appears, check that the connector is switched on for that chat and that you connected it. If the assistant sends text without a review link, that is the limit described at the top: it has no hook, so nothing stopped it.