AI Text Check in ChatGPT
Flag AI-writing tells in text. The MCP server is https://aitext.openkrill.app/mcp. No key. The steps below are only for ChatGPT. A call to check_ai_writing was checked against that server on 2026-10-01.
What this server answers
Paste a draft and see the habits that make writing read like it came from an AI model, each marked at its exact place with a plain-language fix.
Ask "which parts of this email sound like AI?", "check my blog post for filler" or "does this cover letter lean on stock phrases?".
AI Text Check looks for overused words and stock openers, hedging openers, sycophantic praise and chatbot sign-offs, generic closers, inflated claims, unnamed experts, "not just X, it's Y" turns, lists of three, heavy em-dash use, decorative bold bullets, and runs of sentences with the same length or the same opening.
Every finding gives the rule, a severity, the start and end offsets, the matched words, why it reads that way and a concrete fix. A score from 0 to 100 sums them up, and the answer states how the score is worked out.
The rules are fixed patterns, not an AI model, so the same text always gives the same answer. They are written for this service; the categories follow the open-source blader/humanizer catalogue (MIT licence) and Wikipedia's "Signs of AI writing".
A finding is a place to look at, not proof of who wrote the text: careful writers use all of these on purpose, and the tool is not an AI detector. It does not rewrite the text.
The text is checked and forgotten: it is never stored, and the only lasting record is a daily count of calls. It is limited to 12,000 characters (about 2,000 words); a score needs at least 40 words.
What it can do
- Flags overused words, stock openers and filler phrases
- Flags hedging openers, generic closers and chatbot sign-offs
- Flags sycophantic praise and inflated claims
- Flags lists of three, heavy em-dash use and decorative bold bullets
- Flags runs of sentences with the same length or opening
- Gives start and end offsets, a severity and a fix for each finding
- Gives a 0 to 100 score and states how it is worked out
- Uses fixed rules, with no AI model, and never stores the text
Tools
check_ai_writing, Check text for AI-writing tells. Flags writing habits. It cannot say who wrote the text. It marks each habit with a fix and a 0 to 100 score. Use it when the user asks which parts read like AI writing, for example "find the filler in my draft", "what makes this email read as machine-written". Pass the text exactly as written, at most 12,000 characters. Returns up to 60 findings in text order, each with a rule id, a severity (high, medium or low), start and end offsets in the text, the matched words, why it reads that way and a plain-language fix, plus a 0 to 100 score with how it was worked out, the source and the date of the rules. The rules are fixed patterns, not an AI model. A finding or a low score is not proof of who wrote the text, so it cannot say whether an AI wrote it, and the tool does not rewrite it. The text is not stored.
Add AI Text Check in ChatGPT
ChatGPT does not read a local config file. A custom MCP server is added in the product, and only after developer mode is on. These steps were checked against OpenAI's plugin connect guide and the ChatGPT MCP guide on 2026-10-01. Developer mode can be hidden by a workspace policy, so a Team or Enterprise admin may have to allow it first.
- Open ChatGPT settings, then Security and login, and turn on Developer mode.
- Open ChatGPT plugins and select the plus button.
- Enter a name such as "AI Text Check" and a one-line description of what you want it to answer.
- Under Connection, choose the public endpoint and paste
https://aitext.openkrill.app/mcp. Include the/mcppath. Do not add a tunnel id: this server is already public. - Create the connection and review the tools ChatGPT lists. You should see
check_ai_writing. The feedback tools on the same server are optional and not required for the lookup. - Start a new conversation, add the connection from the tools menu, and ask: Which parts of this email sound like AI wrote them?
No API key and no OAuth client id are required. The server answers without a login. Calls are limited per network and per day, and a limit comes back as a tool error rather than a partial answer. If the tool list is stale after a server change, open the connection and select Refresh, then start a new conversation.
The ChatGPT desktop app, Codex CLI and IDE extension share a different config. In the desktop app, open Settings, then MCP servers, choose Add server, select Streamable HTTP, and paste the same URL. In ~/.codex/config.toml the equivalent is a table named mcp_servers.ai-text with url = "https://aitext.openkrill.app/mcp". ChatGPT on the web does not read that file. A published directory listing is also a different path: this page is the developer-mode connection, not a claim that AI Text Check is already in the ChatGPT directory.
Treat the server as untrusted until you have read a few answers. A remote MCP server sees the arguments the model sends. AI Text Check is read-only and documents what it sends onward on the privacy page.
A call checked on 2026-10-01
A 59-word draft with stock openers is enough for a score. The tool uses fixed rules, stores nothing, and a score is not proof of who wrote the text. The request below was posted to https://aitext.openkrill.app/mcp as tools/call. HTTP 200. Source: the AI Text Check server, read 2026-10-01.
{
"method": "tools/call",
"params": {
"name": "check_ai_writing",
"arguments": {
"text": "In today's rapidly evolving landscape, it is important to note that this update is not just a change, it's a revolution. Furthermore, the team leveraged cutting-edge synergy to empower stakeholders and unlock seamless potential. Moreover, this robust solution stands as a testament to our commitment to innovation. In conclusion, the future looks bright as we continue this journey together."
}
}
}- as of. 2026-10-16
- score. 100
- level. high
- findings. 11
- source. Deterministic pattern rules written for this service, with no AI model. Rule categories follow blader/humanizer (MIT licence) and Wikipedia's Signs of AI writing.
- words. 59
Repeat the call yourself if you need a newer reading. Cached answers expire. A rate limit is not a result: wait and try again. Nothing in the call is a ranking, a filing, or advice.