Your text should sound like you.

Paste your text, get an indicative style score, and remove formulaic phrasing without losing numbers, links, or required terms.

Your text

Analysis runs locally. Editing uses OpenAI.

02 / METHOD

Not a truth detector. An editor's tool.

The index spans 8–98 on a 100-point display and aggregates repetition, sentence rhythm, stock transitions, and other observable features. It is not an AI-authorship probability and does not prove provenance.

Invariants first

Numbers, links, and required terms are compared before and after editing.

Voice second

The model edits clarity and rhythm without inventing experience or sources.

Always qualified

Short text receives a wider interval and a lower-confidence assessment.

03 / MCP

Connect Human for Human over MCP.

Analyze and rewrite text from Codex, the ChatGPT desktop app, or Claude Code. The server uses Streamable HTTP and is available on this domain.

Server endpoint: /mcp

Connect

Codex and the ChatGPT app

codex mcp add human-for-human --url /mcp

Codex CLI, the ChatGPT desktop app, and the Codex extension share MCP configuration. Restart the client after adding the server, then verify it with /mcp.

Claude Code

claude mcp add --transport http human-for-human /mcp

This adds the server to the current project's local configuration. Open a new Claude Code session after connecting.

Available tools

analyze_text: Shows how formulaic the text is using an 8–98 score, an assessment range, and core metrics. It makes no OpenAI request and does not determine authorship.
rewrite_text: Edits the text for the selected voice, audience, and goal. It checks numbers, URLs, and required terms before returning the draft. OpenAI processes the text with store=false.

Before a public launch, add HTTPS, authentication, and request limits: the endpoint currently requires no authentication.

The score is a heuristic, not an authorship probability. Rewriting does not guarantee passing AI detectors.

Codex MCP documentation