AI Detector MCP for Claude, Codex, and Cursor on Mac

An AI Detector MCP is useful when the check belongs inside the agent session, not in another browser tab. Claude, Codex, Cursor, Hermes Agent, and OpenClaw can ask a local tool to score a draft, inspect an image, check readability, or clean hidden text artifacts while the work is still in progress.

Slop or Not now ships that local tool with the Mac app. The slop mcp server runs as a stdio MCP server, provides the same detection primitives as the app, and keeps the actual checks on your Mac. The bundled slop CLI is still there for Terminal and scripts, but this post focuses on MCP because that is the better fit for agent workflows.

The setup has four steps: install Slop or Not for Mac, activate Pro, use the Command Line tab for the current CLI setup command, then register slop mcp in your MCP client.

What Is an AI Detector MCP?

An AI Detector MCP is a local Model Context Protocol server that gives an agent tool access to AI detection. Instead of pasting text into a detector, copying JSON back, and asking the agent what changed, the agent calls the detector directly and receives structured results.

In Slop or Not, the MCP server is bundled inside the signed Mac app. It runs through the slop mcp command and talks to MCP clients over stdio. The client sends the text or image payload to the local command. Slop or Not runs the detection, readability, cleanup, or image check on your Mac and returns a tool result to the client.

That matters for two reasons. First, it removes the manual copy-paste loop. Second, it keeps Slop or Not's part of the check local. If the agent itself rewrites text through a cloud model, that model's privacy policy still applies to the rewrite. The Slop or Not detection step does not upload the checked text or image to a Slop or Not server.

The MCP server requires Pro on Mac because it is built for repeated agent and automation use. The free app remains the right starting point if you only need manual checks.

How Does Slop or Not MCP Work on Mac?

Slop or Not MCP starts the bundled slop binary and provides six local tools to the MCP client. The server does not create a hosted detector endpoint. It is a local process launched by Claude, Codex, Cursor, Hermes Agent, OpenClaw, or another MCP-capable app.

The binary lives inside the Mac app bundle:

/Applications/Slop Or Not.app/Contents/MacOS/slop

Most clients are easiest to configure with the app-bundle path. Open Slop or Not once, activate Pro, then open Settings > Command Line for the current shell setup command. Current builds use this path:

"/Applications/Slop Or Not.app/Contents/MacOS/slop" status --json

The expected status output includes a version number and Pro state. After that, your MCP client can launch the server with:

"/Applications/Slop Or Not.app/Contents/MacOS/slop" mcp

For GUI agents, use the direct app-bundle command: /Applications/Slop Or Not.app/Contents/MacOS/slop. That keeps MCP setup tied to the signed Mac app instead of the way a client launches its shell.

Which Tools Does the AI Detector MCP Provide?

The AI Detector MCP provides text detection, image detection, readability analysis, text cleanup, image scoring, and install status. These map to the same local checks available in the Mac app and the slop CLI, but MCP returns them as callable tools inside the agent.

MCP tool	What it does
`slop_status`	Checks app install state, version, and Pro access
`detect_text`	Scores pasted text for likely AI generation
`analyze_readability`	Returns reading-level and readability metrics
`clean_text`	Finds and removes hidden characters, homoglyphs, and odd punctuation artifacts
`detect_image`	Checks a local image payload for likely AI generation
`score_image`	Returns the raw OmniAID image score

Text detection is the main authorship check. Readability is separate: it gives you reading-level feedback, not an AI verdict. Text Cleanup is also separate. It does not rewrite prose or "humanize" a draft. It strips mechanical artifacts that should not be part of the text.

Image checks read provenance signals first when supported, including SynthID watermarks plus C2PA and IPTC metadata used by major image generators. If no readable watermark or metadata is available, Slop or Not falls back to the on-device image model.

How Do I Add the AI Detector MCP to Claude, Codex, and Cursor?

Register slop mcp as a stdio server in the MCP client you use. The command is the same in every client, but the config format changes between Claude, Codex, Cursor, Hermes Agent, and OpenClaw.

Claude Code can add the server from Terminal:

claude mcp add --transport stdio --scope user SlopOrNot -- "/Applications/Slop Or Not.app/Contents/MacOS/slop" mcp

Codex uses TOML in its MCP server config:

[mcp_servers.SlopOrNot]
command = "/Applications/Slop Or Not.app/Contents/MacOS/slop"
args = ["mcp"]

Cursor uses JSON, either globally or per project:

{
  "mcpServers": {
    "SlopOrNot": {
      "command": "/Applications/Slop Or Not.app/Contents/MacOS/slop",
      "args": ["mcp"]
    }
  }
}

Hermes Agent uses an MCP server entry:

mcp_servers:
  SlopOrNot:
    command: "/Applications/Slop Or Not.app/Contents/MacOS/slop"
    args: ["mcp"]

OpenClaw can register the same local command:

openclaw mcp set slopornot '{"command":"/Applications/Slop Or Not.app/Contents/MacOS/slop","args":["mcp"]}'

After registration, restart the client and run whatever MCP inspection command it provides. Claude Code, for example, can verify configured servers with /mcp. The full setup lives in the Slop or Not MCP guide.

When Should I Use MCP Instead of the CLI?

Use MCP when the agent should decide when to call the detector. Use the CLI when you want an explicit shell command, a script, or JSON output you will store outside the conversation.

MCP is better for:

Claude, Codex, Cursor, Hermes Agent, or OpenClaw review loops
agentic AI Humanizer workflows
repeated readability checks during a rewrite
cleaning hidden text artifacts before a second detector pass
checking images without leaving the agent session

The CLI is better for:

batch checks from Terminal
local scripts
one-off image checks
raw OmniAID image scoring with score-image
debugging the app install
piping clipboard text into JSON output

For direct model feedback in Terminal, use the CLI score-image command:

"/Applications/Slop Or Not.app/Contents/MacOS/slop" score-image --json < generated.png

Both surfaces use the same on-device Slop or Not checks. The difference is who calls the tool. With CLI, you call it. With MCP, the agent calls it when the workflow asks for an AI detector, readability analyzer, cleanup pass, or image score.

How Does This Help an Agentic AI Humanizer Loop?

An agentic AI Humanizer loop needs measurement between rewrites. MCP gives the agent a local way to score the baseline, clean mechanical artifacts, rewrite the draft, check readability, and score the result again without asking you to shuttle JSON between tools.

A typical loop looks like this:

The agent calls detect_text to get a baseline.
It calls analyze_readability to record the reading level.
It calls clean_text if copied text includes hidden characters or odd punctuation.
It rewrites the draft using your stated style and constraints.
It calls detect_text and analyze_readability again.
It compares the result and decides whether another pass is useful.

Slop or Not does the local measurement. The agent still owns the rewrite. If you want the rewrite to sound like you, give the agent a writing sample and constraints before the loop starts. The agentic AI Humanizer guide covers the full setup, and the own-voice AI humanizer guide explains how to keep the rewrite from flattening your style.

FAQ

Is the Slop or Not AI Detector MCP free?

The Slop or Not app is free with limited daily checks. The MCP server and CLI require Pro on Mac because they are built for repeated agent and automation workflows.

Does the Slop or Not MCP server upload my text or images?

No. Detection, readability, cleanup, and image checks run locally on your Mac. If your AI agent rewrites text through a hosted model, that hosted model's own privacy policy applies to the rewrite step.

Can I use the AI Detector MCP without Claude?

Yes. Claude is only one MCP client. The same slop mcp server can be registered in Codex, Cursor, Hermes Agent, OpenClaw, and other clients that support stdio MCP servers.

Do I still need the CLI?

Sometimes. MCP fits agent workflows better, but the CLI is useful for Terminal checks, shell scripts, and debugging. You can use both because they call the same on-device Slop or Not app.

Can the MCP server detect AI images as well as AI text?

Yes. The MCP server provides text and image tools. Image checks use SynthID, C2PA, or IPTC provenance when available and fall back to Slop or Not's on-device image model when the file has no readable provenance.

Try It

Install Slop or Not for Mac, activate Pro, and follow the MCP guide. If you want the command-line version too, the CLI guide covers direct Terminal use.

For writing workflows, start with the agentic AI Humanizer guide. For image-focused checks, read the guide to detecting ChatGPT-generated images next.