Integrations

Use WROITER where you already write.

WROITER runs as a tool inside the AI you already use, or as an API in your own pipeline. It reads your text and hands back specific fixes — the same engine this site runs. Pick your setup below. No account, nothing stored.

Claude ChatGPT Any other tool What you get back

Claude

Verified

Add WROITER as a custom connector. Works on Claude Free (one connector), Pro, Max, Team, and Enterprise.

https://api.wroiter.com/mcp
  1. In Claude, open Settings → Connectors → Add custom connector.
  2. Name it WROITER, paste the URL above, and leave authentication blank.
  3. Open a chat, switch the WROITER connector on, and paste a draft (15 words or more).
  4. Ask: “Run this through WROITER and apply the fixes.”

You'll know it worked when Claude calls the wroiter_revise tool and answers with named fixes — what to change, where, and why — instead of a vague rewrite.

Stuck? admin@wroiter.com

ChatGPT

Plus + Developer Mode

Custom connectors live behind ChatGPT's Developer Mode (a beta) and need a paid plan — Plus, Pro, Business, Enterprise, or Edu. Not available on Free.

WROITER is read-only. It reads your text and returns findings; it never writes anything anywhere. ChatGPT's Developer Mode warnings about tools that can change your data don't apply to it.

https://api.wroiter.com/mcp
  1. In ChatGPT, open Settings → Apps & Connectors → Advanced settings and turn on Developer Mode.
  2. Back in Apps & Connectors, choose Add new connector. Name it WROITER, paste the URL, set authentication to None.
  3. Start a chat, switch the WROITER connector on, and paste a draft (15 words or more).
  4. Ask: “Run this through WROITER and apply the fixes.”

You'll know it worked when ChatGPT calls wroiter_revise and returns the same per-instance fixes.

Stuck? admin@wroiter.com

Any other tool

For Gemini, Grok, a script, or a content pipeline — reach the same engine over MCP or plain HTTP.

Any MCP client

Point any MCP-capable client (Claude Desktop, Cursor, an agent) at the same URL. The tool is wroiter_revise.

https://api.wroiter.com/mcp
The HTTP API

For pipelines and anything that isn't MCP. One POST, no key.

POST https://api.wroiter.com/revise Content-Type: application/json { "text": "your draft — 15 words or more" }

It returns { payload, markdown } — payload is the structured findings, markdown is a revision prompt you can paste into any AI. Up to 100,000 characters, about 20 requests a minute.

Teams drop this between drafting and review as an automatic de-slop pass — the editor opens a cleaned draft instead of catching slop by eye.

Gemini, Grok, Perplexity and more get their own walkthroughs as we verify each one. Until then, the API reaches them all.

What you get back

However you connect, the answer is specific.

Paste a draft and WROITER returns one instruction per problem — the phrase, the fix, and the reason. Here's a real response, with three of its flags.

AI-typical phrase template BANNED_PHRASES high confidence

Findit's important to note

ActionDelete "it's important to note" and start the sentence with the substantive claim that follows. If the sentence has nothing left after the deletion, the sentence had nothing to say.

WhyTemplate phrases defer commitment to a specific claim. Detectors flag them at the lexical level and editors flag them at the reading level; removing them sharpens both at once.

Credibility-asserting framing CREDIBILITY_THEATER high confidence

FindLet me be direct

ActionDelete the framing and make the direct statement itself. Honest writing doesn't announce its honesty; the label only advertises the gap.

WhyPermission-asking and credibility labels are persona leakage — the writing performing a posture instead of carrying the point.

Sequence-marker scaffolding OVER_SIGNPOST high confidence

FindFirst

ActionRemove "First". The reader can see this is the next paragraph; they don't need a label that says so. If the order matters, the content carries it.

WhyStacked sequence markers ("First … Second … Finally") signal essay scaffolding that survived into the final draft.

…and the rest, grouped by severity. Every flag carries a confidence badge — high means the fix is mechanical, medium means it's your editorial call.

How it works
  • No LLM call and no model spend — the engine is deterministic pattern-matching, not a guess.
  • Nothing is stored. Your text is analyzed and dropped; it never persists.
  • Same text in, same findings out, every time.
  • It's the exact engine running on this site, behind a URL.
The full method is published →
Try it in your browser → Read the method