Check Your Text With an AI Detector From Inside Your Chat

Before anyone submits AI-assisted writing, they run the same ritual: copy the text, open ZeroGPT, paste, check the score, open GPTZero, paste again, check again, maybe try Copyleaks if they're nervous. It's tedious, and you never quite trust any single number.

The SupWriter MCP server folds a first check into one tool call. Your AI assistant runs it and gets back an overall AI score, a sentence-by-sentence breakdown, and a row of 12 labels named after common checkers — without you leaving the conversation or pasting into a separate site. One thing to know up front: all of that comes from one scoring model. The labels are a guide mapped from that single score, not twelve separate detectors.

Why a tool call beats the manual ritual

The manual approach has two problems. First, it's slow — every detector is a separate tab and a separate paste. Second, detectors disagree with each other, so no single number settles anything. We've written before about why AI detectors give different results on the same text.

A tool call fixes the first problem, not the second. The detect_ai tool runs one scoring model and returns:

• an overall AI score from 0–100 (higher means more likely AI),
• 12 guide labels named after Turnitin, GPTZero, Copyleaks, ZeroGPT and other checkers, each marked pass / warn / fail. They are filled from that one score using a fixed table (how the mapping works); none of those vendors is queried,
• a sentence-level breakdown so you know which sentences are pushing the score up, instead of guessing.

That last part is the useful one. A single number tells you there may be a problem; the sentence breakdown tells you where to look. If your school or client uses a specific detector, that detector's result is still the one that counts.

Reading the output

When the assistant runs detect_ai, it gets back the score, the label map, and the flagged sentences. A good assistant will summarize it for you — "scores low overall, but these three sentences in the second paragraph are flagged as likely AI" — and you can act on exactly those. Then you revise or humanize the flagged passage, re-run the check, and see how the score moves. Detect → revise → re-detect is a tight loop when it all happens in one thread.

Free vs paid — stated plainly

Over MCP, detection is metered on every plan: each check costs 1 credit per word you send, drawn from the same pool the other tools use. (Detection inside the SupWriter web app is free on paid plans; the MCP server and the API are metered.) Your per-request word cap still applies: 500 words on Free and Basic, 2,000 on Pro, 5,000 on Ultra. If you're out of credits, the tool returns an error that says so instead of a result. Plans are on the pricing page.

An honest caveat about "scores"

No detector is ground truth. They're probability estimates, they disagree, they produce false positives on genuinely human writing, and vendors update their models regularly. A SupWriter detection score is one model's estimate — most useful for its sentence breakdown — not a verdict, and not a forecast of what Turnitin or GPTZero will say. Use it to find weak spots and to sanity-check before you submit, not as proof of anything. If you want to understand the limits, are AI detectors even accurate? is worth your time.

Pair it with the humanizer

The real workflow is the loop: detect to find the flagged sentences, humanize or rewrite them, detect again to compare. Because both tools live in the same SupWriter connector, your assistant can run the whole cycle in one conversation. Each call spends credits, so check the short passages that were flagged rather than re-running the whole document every time.

Connecting it

Add https://supwriter.com/api/mcp in your AI client's connector settings, sign in with your SupWriter account, and approve. Step-by-step for Claude, ChatGPT, and Cursor / Claude Code / VS Code.