Check writing for AI without leaving your assistant
Add ScanForAI to the AI assistant you already use, and it can check any text for signs of AI writing and tidy its typography right in the chat. It works through MCP, the standard way assistants connect to outside tools. A score shows you where to look closer. It is never proof of who wrote something.
Add it to your assistant
You need a key from a free ScanForAI account. It takes a few minutes and no card.
- Create a free account.
Sign up and confirm your email address from the message we send you.
- Make a key.
Open the API keys page of your account and create a key. Set Scope to Full so both tools work. The key starts with
at_sk_. Keep it private, like a password. - Add it to your assistant.
Copy the lines for your assistant below and put your own key where they ask for it.
Claude Code
Run this in a terminal. Claude Code saves it for all your projects, so you only do this once.
claude mcp add --transport http --scope user scanforai \ https://api.scanforai.com/api/mcp \ --header "Authorization: Bearer at_sk_your_key_here"
Cursor
Add this to ~/.cursor/mcp.json to use it in every project, or to .cursor/mcp.json inside one project. Keep your key out of files you share.
{
"mcpServers": {
"scanforai": {
"url": "https://api.scanforai.com/api/mcp",
"headers": {
"Authorization": "Bearer at_sk_your_key_here"
}
}
}
} Codex CLI
Add this to ~/.codex/config.toml. Codex sends the key that is stored in the variable it names.
[mcp_servers.scanforai] url = "https://api.scanforai.com/api/mcp" bearer_token_env_var = "SCANFORAI_API_KEY"
Then set that variable with this line before you start Codex. To keep it set, add the line to your shell profile, for example ~/.zshrc. On Windows, set it as an environment variable instead.
export SCANFORAI_API_KEY="at_sk_your_key_here"
VS Code with GitHub Copilot
Open the Command Palette, run MCP: Open User Configuration and add this. VS Code asks you for the key and hides it as you type.
{
"inputs": [
{
"type": "promptString",
"id": "scanforai-key",
"description": "ScanForAI API key",
"password": true
}
],
"servers": {
"scanforai": {
"type": "http",
"url": "https://api.scanforai.com/api/mcp",
"headers": {
"Authorization": "Bearer ${input:scanforai-key}"
}
}
}
} Claude Desktop
Claude Desktop connects through a small helper called mcp-remote, which needs Node.js on your computer. From the Claude menu, open Settings, then Developer, then Edit Config, and add this:
{
"mcpServers": {
"scanforai": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://api.scanforai.com/api/mcp",
"--header",
"Authorization:${AUTH_HEADER}"
],
"env": {
"AUTH_HEADER": "Bearer at_sk_your_key_here"
}
}
}
} Save the file, then quit Claude Desktop completely and open it again.
Another assistant
Any assistant that can add a remote MCP server with a request header can use ScanForAI. Give it this address and this header:
https://api.scanforai.com/api/mcp Authorization: Bearer at_sk_your_key_here
What the two tools do
scan_text: check for signs of AI
Scores a text from 0 to 100 for how strongly it shows signs of AI writing, gives a band from low to high, a short reading and the names of the signs it found, which your assistant can explain. Ask for the thorough read when it matters: it is slower and comes from a smaller allowance. Text in a language we cannot read yet gets no score, and the answer says so.
cleanup_text: tidy the typography
Turns the curly quotes, long dashes and rare symbols that AI tools tend to leave behind into plain keyboard characters, and gives the text back. It does not change a single word. It is not a humanizer: it removes formatting tells, not the writing patterns a detector weighs.
You do not need the tool names. Ask in your own words, for example “check this paragraph with ScanForAI” or “clean up the quotes in my draft”, and your assistant usually picks the right tool.
A score is a signal, not a verdict. Students: a high score on your own writing is not proof against you, and the signs it names show what to look at. Teachers: treat a score as a reason to talk with the student, never as evidence on its own.
Keys, limits and your plan
- A free account is enough. Every scan your assistant runs counts against your plan, the same as a scan on the site or through the API.
- You need a confirmed email to make a key. Confirm it from the message we send when you sign up.
- New accounts start smaller. A free account made from an MCP sign-up link, such as the ones on this page, has a smaller allowance for the thorough read during its first month, on the site too. Upgrading lifts it.
- Reaching a limit. Your assistant gets the quick read instead, or is told to try again later. See pricing for what each plan includes.
What we keep
Scans and clean-ups made with your key are saved to your account history, just like in the web app. You can delete them there, or delete your whole account, at any time.
With each call we also note which assistant made it, using the name the assistant sends, such as Claude Code or Cursor.
Your assistant reads the text before it sends it to us, so your assistant's own privacy rules apply on its side. There is no anonymous use through MCP: every call needs a key. Our privacy policy has the details.
Not through MCP yet
Through MCP, ScanForAI checks and tidies text. These other checks run in the web app or through the REST API today:
- Hidden instructions in images. Finds writing inside a picture that an AI would read and you would not. Use it in the web app or the REST API, and see how it works.
- How a picture was made. Reads what an image file says about the tool that made it. Available through the REST API.
- Files and web pages. Upload a document or paste a web address on the site or in the web app. Files also work through the REST API.
Questions about using it from an assistant
Is it free?
Yes, with a free account. Your assistant's scans count against your plan like any other scan, and paid plans raise the limits. See pricing.
Which assistants does it work with?
Claude Code, Cursor, VS Code with GitHub Copilot, Codex CLI and Claude Desktop, set up as shown above. Any other assistant that can add a remote MCP server with a request header works too. As a custom connector in Claude on the web or in the desktop app, it works only where Anthropic has turned on request headers, which only some accounts have today, and only with your own key on an account only you use.
Can I use it without an account?
No. Every call through MCP needs a key, and a key needs a free account. To check a text without signing up, paste it into the scanner on the site.
Does my assistant keep the text I check?
Your assistant reads the text before it sends it, so what it keeps is up to its maker and your settings there. On our side, the scan is saved to your account history, and you can delete it at any time.
Why did I get no score?
Usually because the text is in a language we cannot read yet. A Spanish text also gets no score when it is too short, or when your thorough reads are used up for now. Either way, the answer says why instead of guessing a number.
Will cleanup help AI writing get past a detector?
No. It removes formatting tells, not the writing patterns a detector weighs. Cleanup swaps machine typography, such as curly quotes and long dashes, for plain characters and changes no words. It is not a humanizer.
Does it work in Spanish?
Yes. English and Spanish texts are scored today. If you want the answer written in Spanish, ask your assistant to set the language to Spanish (es).
Where is the ScanForAI MCP server listed?
In the official MCP registry as com.scanforai/scanforai, on Glama, and on GitHub. To call ScanForAI from your own code instead, see the REST API docs.