Introducing the AuraWrite MCP Server: Humanize and Detect AI Text Inside Claude and ChatGPT
You draft in Claude or ChatGPT, then copy the text into AuraWrite, then copy it back. That loop is gone. Connect our MCP server once, and your assistant can humanize, detect, grammar-check, and paraphrase for you, right in the conversation.
Most people who use AuraWrite don't start in AuraWrite. They start in a chat window. They ask Claude for a first draft of a cover letter, or have ChatGPT turn meeting notes into a blog post, and only once the draft exists do they open a second tab, paste it into our humanizer, wait, copy the result, and paste it back into the chat to keep working. For a single paragraph that's a minor annoyance. For a ten-section report you're revising all afternoon, it's dozens of round trips and a real chance of pasting the wrong version somewhere.
The AuraWrite MCP server removes the second tab. It exposes our humanizer, AI detector, grammar checker, and paraphraser as tools your AI assistant can call directly. You stay in the conversation, describe what you want, and the assistant handles the hand-off. This post covers what MCP is, exactly which tools are available and what their limits are, how to connect in Claude, ChatGPT, and Claude Code, how billing works, and the workflows where it saves the most time.
1. What Is MCP, in Plain Terms?
MCP stands for Model Context Protocol. It's an open standard that lets AI assistants use outside tools in a consistent way. Think of it as a universal plug: a service like AuraWrite publishes a list of tools it offers, what each tool expects as input, and what it returns. Any assistant that speaks MCP can then read that list and decide when to call a tool on your behalf.
The practical upshot is that you don't have to learn any commands. You don't type humanize_text anywhere. You say "humanize this," or "check whether this reads as AI," and the assistant figures out that the AuraWrite tool is the right one, sends it your text, and shows you the result. If you ask for something that needs two steps, like detecting first and humanizing only if the score is high, the assistant can chain the calls itself.
Our server is a remote MCP server, which means nothing gets installed on your computer. It lives at a single URL, https://aurawriteai.com/api/mcp, and your assistant talks to it over the internet using the Streamable HTTP transport. You sign in once with your AuraWrite account, and from then on the connection just works.
2. The Five Tools You Get
The MCP server ships with the same core tools you use on the website, plus one for checking your balance:
| Tool | What it does | Limit |
|---|---|---|
| humanize_text | Rewrites AI-generated text so it reads like a person wrote it. | Up to 3,000 words per request |
| detect_ai | Scores text from 0 (human) to 100 (AI) with a sentence-by-sentence breakdown. | 50–5,000 characters |
| check_grammar | Fixes grammar, spelling, and punctuation without rewriting your voice. | Up to 5,000 characters |
| paraphrase_text | Rewords text with new phrasing while keeping meaning, tone, and length. | Up to 5,000 characters |
| get_usage | Shows your plan, words used, words remaining, and reset date. | Free to call |
A few details worth knowing. The humanizer returns only the rewritten text, with no preamble, so the assistant can drop it straight into whatever you're building. The AI detector returns an overall score plus a score for each sentence, which means you can ask follow-up questions like "which three sentences are pulling the score up?" and get a specific answer instead of a single number.
The grammar checker is deliberately conservative. It fixes errors and leaves your phrasing alone, so it's safe to run on text you've already humanized without undoing that work. The paraphraser does the opposite. It changes wording and sentence structure on purpose while holding the meaning, tone, and rough length steady.
3. How to Connect It (Under a Minute)
You'll need an AuraWrite account. If you don't have one, you can create one for free. There's no API key to generate or paste; every app below signs you in through a secure OAuth login using your normal AuraWrite credentials.
Claude (claude.ai and Claude Desktop)
- Open Settings and go to Connectors.
- Click Add custom connector.
- Paste
https://aurawriteai.com/api/mcpand save. - Click Connect and sign in to your AuraWrite account.
ChatGPT
- Open Settings and go to Apps & Connectors.
- Under Advanced, turn on Developer mode.
- Click Create and paste the AuraWrite MCP URL.
- Set authentication to OAuth, then sign in to your AuraWrite account.
Claude Code
Run this in your terminal:
claude mcp add --transport http aurawrite https://aurawriteai.com/api/mcpThen start Claude Code, run /mcp, and sign in. This one is especially handy if you write documentation, changelogs, or README files alongside your code. You can have the copy humanized and grammar-checked without leaving your editor.
Any other MCP client
If your tool supports remote MCP servers with Streamable HTTP and OAuth sign-in, point it at the same URL. The full setup guide, with a copy button for the URL, lives on our AuraWrite MCP page.
4. What It Costs
Nothing extra. The MCP server draws from the same monthly word balance as your AuraWrite plan. There isn't a separate MCP subscription, and there isn't a second pool of credits to track. Words you use in Claude, ChatGPT, or Claude Code and words you use on aurawriteai.com come out of one shared balance.
- One word credit per input word. Humanizing, detecting, grammar-checking, and paraphrasing each count the words in the text you send.
- You're only charged when a request succeeds. The server checks your balance first, runs the tool, and deducts words only after the result comes back.
- Checking your balance is free. Ask "How many AuraWrite words do I have left?" as often as you like.
- Clear errors when you run low. If a request needs more words than you have, the assistant tells you how many it needed, how many remain, and when your plan resets, instead of failing silently.
Free accounts work over MCP too, which makes it easy to try the connection before committing to a plan. If you outgrow the free allowance, the paid tiers on our pricing page apply everywhere at once. Upgrade on the website and your assistant picks up the new limit on its next call.
5. Workflows That Get Noticeably Faster
The value of MCP isn't any single tool call. It's that the assistant can combine them with everything else it's already doing. Here are the patterns we've seen save the most time.
Draft, then humanize in one breath
Instead of drafting and then running a separate humanizing pass, ask for both at once: "Draft a LinkedIn post about our product launch, then humanize it before showing me." The assistant writes the draft, sends it through humanize_text, and shows you only the finished version. You still review it, but you skip the copy-paste loop entirely.
Detect first, humanize only what needs it
Not every paragraph needs rewriting, and rewriting text that already reads as human just spends credits. Try: "Check this essay for AI and humanize anything over 50%." The assistant runs the detector, reads the sentence-level scores, and sends only the flagged sections through the humanizer. On a long document this is both faster and cheaper than humanizing everything.
Humanize, then polish
"Humanize this cover letter so it sounds like me, then fix any grammar mistakes." Two tools, one request. Because the grammar checker only corrects errors and doesn't rephrase, the natural voice from the humanizing pass survives intact.
Long documents, section by section
The MCP humanizer accepts up to 3,000 words per request, which covers most blog posts and essays in one go. For anything longer, like a thesis chapter or a full report, just tell the assistant: "Humanize this report one section at a time and keep the headings exactly as they are." It will split the text, process each part, and stitch the result back together, which also keeps each section's structure cleaner than one giant pass would. If you'd rather process a very long piece in a single request, the website humanizer accepts up to 6,000 words.
Tone adjustments without starting over
"Paraphrase this paragraph in a more casual, friendly tone." The assistant can use the paraphraser for the structural rewrite and then adjust tone around it, which is often quicker than re-prompting a draft from scratch and hoping the new version keeps the facts straight.
6. What a Session Actually Looks Like
Here's a typical exchange once AuraWrite is connected:
YOU
Here's the intro to my product update email. Run it through the AuraWrite detector and tell me which sentences sound the most like AI.
ASSISTANT
The intro scored high overall. Two sentences carry most of it: the opener ("We're thrilled to unveil a game-changing update…") and the closing line about "empowering teams to unlock their full potential." The middle sentence with your actual release date and feature names scored as human. Want me to humanize just those two?
YOU
Yes, and keep the release date exactly as written.
That's the whole interaction. No tab switching, no pasting, and the assistant can explain the detector's output in context instead of leaving you to interpret a percentage. Notice also that it pointed out the specific sentence that already read as human, the one with real dates and names. That pattern holds almost everywhere: concrete details read as human, and generic enthusiasm reads as AI.
7. Privacy and Account Security
Because the connection uses OAuth, your AuraWrite password is never shared with Claude, ChatGPT, or any other client. You sign in on our login page, and the app receives a token that lets it call AuraWrite tools on your behalf. Every request is verified against that token and tied to your account, so your word balance can only be spent by apps you've connected yourself.
You can disconnect at any time. Remove the AuraWrite connector from the Connectors or Apps section of your Claude or ChatGPT settings, or run claude mcp remove aurawrite in Claude Code. The text you send is processed the same way as text you paste into the website. See our privacy policy for the details.
8. Tips for Getting the Best Results
- Name the tool when it matters. Assistants usually pick the right tool, but saying "use AuraWrite to humanize this" removes any doubt, especially if you have other writing connectors installed.
- Tell it what to protect. Quotes, statistics, product names, citations, and required legal wording should stay exact. Say so up front, and ask the assistant to compare those details against the original after the rewrite.
- Detect before you humanize long pieces. It costs credits too, but on a long document it usually saves more than it spends, because you only rewrite the sections that actually need it.
- Re-check after big edits. If you rewrite a section heavily by hand after humanizing, a quick detector pass confirms it still reads naturally.
- Keep your own voice in the loop. The best results still come from drafts that contain things only you know: the specific number, the real anecdote, the actual reason. The humanizer smooths the rhythm, and your details make it credible.
Bring AuraWrite into your AI assistant
Create a free account, paste one URL into Claude or ChatGPT, and humanize, detect, and polish without leaving your chat.
Conclusion
AI assistants have become where a lot of writing actually happens, and the tools that help you finish that writing should live there too. The AuraWrite MCP server puts our humanizer, AI detector, grammar checker, and paraphraser one sentence away from wherever you're drafting. It works in Claude, ChatGPT, and Claude Code, uses your existing account and word balance, and takes about a minute to set up.
If you've been living in the copy-paste loop, try connecting it on your next draft. Ask your assistant to check the piece, humanize what needs it, and fix the grammar, all in one request. Most people don't go back to the second tab.
Last updated: November 16, 2026