AssetFeed AI Detection

npx add-mcp 'https://api.assetfeed.com/mcp/ai-detection'

AI Content Detector

Paste an English text or a page address and see which parts read as AI-written. Free while in beta.

0 / 3,000 words

Without an account: 5 checks a day, up to 3,000 words each. Sign in to check up to 30,000 words at a time, as often as you like.

MCP

Run the same check from Claude, ChatGPT or Cursor.

Add the server once, then ask your assistant to check a draft. Each tool call uses your AssetFeed credits.

More on the MCP page

Chat apps

Coding tools

Local and open-source

1. 01 Open ChatGPT on the web and go to Plugins.
2. 02 Click + and choose Add custom MCP server.
3. 03 Name it AssetFeed AI Detection, paste the server URL and pick OAuth.

   https://api.assetfeed.com/mcp/ai-detection
   

4. 04
5. 05 Ask, for example: “Is this draft written by AI? [paste the text]”

Works on Business, Enterprise and Edu plans. An admin may have to add or allow the server first. Pro connects with read-only access, and all AssetFeed tools are read-only. You can’t add a server in the phone app.

How it works

Two models score the text one part at a time.

01

It splits the text

We cut your text into parts of roughly 120 to 300 words, along paragraph breaks. A long text can read as AI in one part and as human in another, and you’ll see that.

02

Two models score each part

Each part goes to two open-source classifiers. If both say AI, the part is marked AI; if both say human, it’s human. If they disagree, it’s marked unsure.

03

The percentage

It’s the share of your words that sit in parts marked AI. Unsure parts are shown on their own and don’t count toward it.

Limits

What the result can’t tell you

Short texts

Under 50 words there’s too little to go on, so the tool won’t run. A few hundred words give a steadier answer than one paragraph.

English only

The models were trained on English. Other languages aren’t checked.

Mistakes both ways

Formal, formulaic or heavily edited writing by a person can get flagged. AI text that someone rewrote by hand can get through.

Not proof

Treat the result as a reason to read more closely. Don’t use it on its own to judge a person.

Frequently asked questions

01

What is an AI content detector?

A tool that estimates whether a text was written by a person or by a language model. People run essays, articles and freelance copy through one, or their own drafts before they publish. This one shows which parts of your text read as AI, not just one score for the whole page.

02

Is it free?

Yes, while it’s in beta. Without an account you get 5 checks a day, up to 3,000 words each. With an account you can check up to 30,000 words at a time, with no daily limit.

03

How does it work?

It splits the text into parts of about 120 to 300 words, and two open-source classifiers score each one. A part is marked AI or human when both models agree, and unsure when they don’t. The percentage is the share of words in parts marked AI.

04

Which languages does it support?

English only for now. If you paste another language, the tool says it can’t check it.

05

How much text do I need?

At least 50 words, and a few hundred is better. Under 120 words the whole text gets one score, so you won’t see a split by passage.

06

How accurate is it?

No detector gets it right every time. It can flag a person’s writing as AI and miss AI text that was edited. Use the result as an estimate and read the marked passages yourself.

07

Can I use the result as proof that someone used AI?

No. It shows how the text reads to two models. It doesn’t prove who wrote it, and it shouldn’t be the only reason to accuse or penalise anyone.

08

Can I use the detector from my AI assistant?

Yes. The same detector is an MCP tool for Claude, ChatGPT and other MCP apps. The MCP page has the setup steps.