How to add a humanizer to Claude, ChatGPT, Cursor and Claude Code (MCP setup)

One page, every client: the command or menu path for each agent, how sign-in works, and what to do when it doesn't connect.

To add a humanizer to any AI agent, add its MCP server. Zavus's server is at https://api.zavus.ai/mcp. For Claude Code, run claude mcp add --transport http zavus https://api.zavus.ai/mcp; for Claude, add it as a custom connector; for ChatGPT, add a custom MCP server from Plugins; for Cursor, add it to your mcp.json. Then sign in in your browser when the agent asks.

What does a humanizer MCP server do?

It gives your agent a tool it can hand its finished writing to. The Model Context Protocol (MCP) lets AI agents like Claude, ChatGPT, and Cursor use external tools. The Zavus MCP server provides them with a humanize tool, which sends writing from an agent to a separate model to be rewritten. The agent can then check the rewrite against its original draft before presenting it to you.

This is different than using a prompt or a skill, which ask the same model to edit its own writing. See why your AI can't fix its own writing and how to make AI writing sound human for more information.

A few things are the same in every client:

  • The server URL is https://api.zavus.ai/mcp. If the client asks for a name, use zavus.
  • You don't need an API key to use this tool. The first time you use it, it will ask you to sign in to Zavus in a browser.
  • You will need to have a plan or trial. New accounts get a 5-day free trial of the full plan. We require a card up front, but you can cancel before the 6th day and not be charged. If you haven't started your trial yet, the humanize tool will return a link for you to do so.
  • Note that the humanize tool will only accept 40 to 10,000 words at a time. If you pass fewer than 40 words, it will return the text unchanged.

Is there a one-step way to set it up?

Yes, if you have an AI agent that can modify its own settings (for example Claude Code, Codex, or Cursor's agent). All you need to do is paste this prompt and let the agent handle it for you:

Set up Zavus for me. Add the MCP server at https://api.zavus.ai/mcp and name it "zavus". I'll sign in when it asks. Once it's connected, humanize a short paragraph (Zavus needs at least 40 words) to check it works. From then on, run finished writing meant for other people, like emails to clients or customers, posts and announcements, through Zavus's humanize tool before you give it to me.

Make sure to include the last sentence so it knows when to use this tool, and you don't have to ask every time. If you prefer to do this manually, or you're using a client that can't modify its own settings (such as Claude or ChatGPT through their web interfaces), follow the steps below.

How do I add a humanizer to Claude (web and desktop)?

You can add Zavus as a custom connector. If you add it while using claude.ai it will also work within the desktop app, since both use your account.

On a Free, Pro, Max or Ultra plan:

  1. Go to Customize, then Connectors.
  2. Click Add custom connector.
  3. Enter the name zavus and the URL https://api.zavus.ai/mcp. If the dialog asks about authentication, choose a sign-in option, not No sign-in.
  4. Click Add, then Connect, and sign in to Zavus in the window that opens.

The Free plan allows one custom connector. On a Team or Enterprise plan, only an Owner can add one. They go to Organization settings, then Connectors, click Add, then Custom (choosing Web if asked), and enter the URL. Members then go to Customize, then Connectors, find zavus (labelled Custom) and click Connect.

You can turn the connector on or off for a chat with the + button, under Connectors.

How do I add a humanizer to ChatGPT?

To connect, add it as a custom MCP server from the Plugins section on ChatGPT's web version. (Note that the feature has gone by a number of names, including connectors, apps, and plugins, so some guides send you to menus that no longer exist.)

  1. Open chatgpt.com/plugins.
  2. Select the + button, then Add custom MCP server.
  3. Enter a name (Zavus) and a short description.
  4. Under Connection, enter https://api.zavus.ai/mcp.
  5. Under authentication, choose OAuth, read the risk warning and select I understand and want to continue.
  6. Select Create as a plugin, check that the humanize tool appears, and sign in to Zavus when asked.

What you can connect depends on your plan. According to OpenAI's help center, only ChatGPT Business and Enterprise/Edu can use full custom MCP server connections via the web. ChatGPT Pro users can connect custom MCP servers with read and fetch permissions only. In ChatGPT Business, only admins or owners can create custom connections for a given workspace. ChatGPT Plus isn't listed. Note that the humanize tool is marked read-only, but OpenAI has tweaked these rules a few times in 2026, so check out OpenAI's article on custom MCP servers for the latest on what's available for your plan before you count on it.

If you'd rather not use MCP in ChatGPT, check out our guide on how to use custom instructions to make ChatGPT more human.

How do I add a humanizer to Claude Code?

Run one command, then sign in from inside Claude Code.

claude mcp add --transport http zavus https://api.zavus.ai/mcp

Then type /mcp in Claude Code, choose zavus from the list and follow the sign-in steps in your browser. If you prefer the command line, claude mcp login zavus does the same thing from any shell.

By default, this adds the server to your current project only. To make it available in every project, add --scope user:

claude mcp add --transport http --scope user zavus https://api.zavus.ai/mcp

You'll also get a new slash command, /mcp__zavus__humanize, which humanizes the last thing Claude wrote. To have Claude Code use Zavus without being asked, add a line like this to your ~/.claude/CLAUDE.md file: "Run finished writing meant for other people through the zavus humanize tool." For more on Claude-specific options, see how to make Claude write like a human.

How do I add a humanizer to Cursor?

Add Zavus to your mcp.json file: ~/.cursor/mcp.json for all your projects, or .cursor/mcp.json in a project root for that project only.

{
  "mcpServers": {
    "zavus": {
      "url": "https://api.zavus.ai/mcp"
    }
  }
}

Cursor takes care of the OAuth sign-in. Open Cursor Settings, find zavus in the list of MCP servers, click Connect and sign in. You can also skip the file: the Add to Cursor link on your Zavus dashboard fills all of this in for you.

How do I add a humanizer to Codex?

Run this in your terminal:

codex mcp add zavus --url https://api.zavus.ai/mcp

This should open a sign-in page in your browser. If it doesn't, run codex mcp login zavus. On a machine with no browser, add --no-browser, open the link it prints on another machine, and paste the redirect URL back into the terminal. The server is saved in ~/.codex/config.toml, and codex mcp list shows it.

How do I add a humanizer to Gemini CLI?

First, make sure you're using the right tool. As of June 18th 2026, if you're using Google AI Pro/Ultra or the free tier of Gemini Code Assist for Individuals, Google have moved you from Gemini CLI to Antigravity CLI. If you're using a Standard or Enterprise licence of Gemini Code Assist, or a paid API key, Gemini CLI will still work for you.

In Gemini CLI, run:

gemini mcp add --transport http zavus https://api.zavus.ai/mcp

Then start Gemini CLI and run /mcp auth zavus to sign in. The OAuth flow needs a local browser, so it won't work on a headless server.

In Antigravity CLI, add this to your ~/.gemini/config/mcp_config.json file (or .agents/mcp_config.json in a workspace):

{
  "mcpServers": {
    "zavus": {
      "serverUrl": "https://api.zavus.ai/mcp"
    }
  }
}

Then open Agent settings, go to Customizations and click Authenticate. Finish signing in in your browser, and if it gives you a code, paste it back into settings. Antigravity is changing fast, so if these steps don't match what you see, check its MCP docs, or fall back on the one-step prompt above.

How do I add a humanizer to Windsurf?

As of June 2nd 2026, Windsurf has been renamed to Devin Desktop, and their docs are halfway through being migrated. To save you the hassle of finding the right file, you can just paste the one-step prompt above directly into the agent and let it find the file. If you'd rather do it by hand, the Devin docs give this entry for a remote server:

{
  "mcpServers": {
    "zavus": {
      "url": "https://api.zavus.ai/mcp",
      "transport": "http"
    }
  }
}

Where it goes depends on your version. Devin's MCP docs say ~/.config/devin/mcp_config.json (%APPDATA%\devin\mcp_config.json on Windows), or .devin/mcp_config.json in a project. The Devin Desktop FAQ still lists the old Windsurf path, ~/.codeium/mcp_config.json. To see which servers the app has loaded, open the Agents window, click Customize and open the MCPs tab. When zavus shows Needs authentication, click Authenticate and sign in. If none of this matches your screen, check Devin's MCP docs.

How do I add a humanizer to GitHub Copilot in VS Code?

Run this in your terminal:

code --add-mcp '{"name":"zavus","type":"http","url":"https://api.zavus.ai/mcp"}'

That adds the Zavus server to your VS Code user profile. You can also run MCP: Add Server from the Command Palette, choose HTTP, paste in the URL and give it the ID zavus. Then start the zavus server from the list of MCP servers and sign in when VS Code asks. Copilot's agent mode can then use the humanize tool.

How do I check it's working?

To check it's working, just ask your agent to "humanize" something. It needs to be at least 40 words, the minimum Zavus will rewrite. For example, you could ask: "Write a 60-word note to a client confirming Thursday's call, then run it through Zavus." You should see your agent make a function call to the humanize tool and then give you back the humanized text.

The first time you do this, your agent should also explain how Zavus works from then on: it runs on finished writing that will be used by other people, you can ask for it any time (just say "humanize this" or "run it through Zavus" to your agent), and you can tell it to always or never do this. You can ask how many words you have left at any time ("how many Zavus words do I have left?") and your agent should call the usage tool. This is free.

What if the humanizer doesn't connect or work?

Most problems are one of these:

Your agent can't see the tool. In most cases, restarting the client or starting a new chat will fix this, since most agents only load their tools when a new chat starts. For agents like Claude and ChatGPT, make sure you have the connector or plugin turned on for the chat.

It never asked you to sign in, or sign-in failed. Start it by hand: /mcp in Claude Code, codex mcp login zavus in your terminal for Codex, /mcp auth zavus in Gemini CLI, or Connect in Cursor's MCP settings. In Claude Code, if the sign-in page opens in your browser but never gets back to the app, copy the full URL from the browser's address bar and paste it into the prompt.

Your agent is telling you to sign in again, even though you already did before. In this case your sign-in has probably expired, and you just need to sign in using the same method you used before.

You can't find the option in ChatGPT. This is most likely related to your ChatGPT plan or your workspace giving you access, so check out the ChatGPT section above.

You can't add the connector in Claude. If you're on the Free plan, you might already have a custom connector added (you can only have one). If you're on Team or Enterprise, you need to have an Owner on your workspace add it for you.

"Zavus needs at least 40 words." The text is too short to rewrite, so leave it as it is.

"Send at most 10,000 words per request." Split the text into parts and humanize each one separately.

Your agent is telling you about a trial or to resolve a billing issue and giving you a link. This means you either haven't started your free trial yet, you've used up the words in your trial, or there's an issue with a payment. Your agent should tell you what the issue is, and you can resolve it by clicking the link. If you're part of a workspace, you'll need to ask an admin to do this since only workspace admins can manage billing.

"The model is starting up." Give it a couple of minutes and try again.

Your agent won't use Zavus unless you tell it to. You'll want to tell your agent when to use Zavus by adding it to your standing instructions: these might be in a CLAUDE.md, AGENTS.md, in your Instructions for Claude, or as a custom instruction in ChatGPT.

For anything else, Zavus support can help, and the Zavus docs cover the API.

Can I use a humanizer without MCP?

Yes, through the REST API. Send text to POST https://api.zavus.ai/v1/humanize with an API key (keys start with zv_). API keys come with the Team plan, where unlimited teammates share 300,000 words a month. That route suits agencies and product teams building Zavus into their own tools; Zavus for agencies has more.

For most people, using the MCP is easier since you can just keep using your agent as normal and it will send finished drafts to Zavus before showing them to you.

Frequently asked questions

What is a humanizer MCP server?

It's a way for your AI agent to use a rewriting model through the Model Context Protocol. The agent writes the text, sends it to the server, where another model rewrites it, then reviews the rewrite and gives it to you.

Do I need an API key to use Zavus over MCP?

No. The first time you use Zavus in an agent that supports MCP, it asks you to sign in through your browser. API keys are only for the REST API, and they come with the Team plan.

Why does Zavus say my text is too short?

Zavus needs at least 40 words to rewrite text. Anything shorter, like a one-line reply, is better left as it is.

Can I use a humanizer MCP server in ChatGPT Plus?

OpenAI's help center lists full custom MCP connections for ChatGPT Business and Enterprise/Edu on the web, with Pro limited to read and fetch actions. Plus isn't listed, so check OpenAI's current help article for your plan.

Sources

  1. Add a connector that isn't in the directory, Claude Docs, accessed November 2026
  2. Connecting custom MCP servers in ChatGPT, OpenAI Help Center, accessed November 2026
  3. Connect and test your plugin, OpenAI Developers, accessed November 2026
  4. Connect Claude Code to tools via MCP, Claude Code Docs, accessed November 2026
  5. Model Context Protocol (MCP), Cursor Docs, accessed November 2026
  6. MCP in Codex, OpenAI, accessed November 2026
  7. MCP servers with the Gemini CLI, Gemini CLI Docs, accessed November 2026
  8. An important update: Transitioning Gemini CLI to Antigravity CLI, Google Developers Blog, May 2026
  9. Model Context Protocol (MCP), Antigravity Docs, accessed November 2026
  10. Devin Desktop FAQ, Devin Docs, accessed November 2026
  11. MCP configuration, Devin Docs, accessed November 2026
  12. MCP configuration reference, Visual Studio Code Docs, accessed November 2026