Connect ReverseGPT to Claude and ChatGPT
One endpoint turns any MCP client into an AI humanizer. Your assistant rewrites AI-generated text as a tool call, using your existing plan and credits.
https://www.reversegpt.ai/api/mcpStreamable HTTP. OAuth 2.1 with dynamic client registration and PKCE, or an API key header — whichever your client prefers.
Tools
Two tools, one job. ReverseGPT does one thing through MCP — it humanizes text — and the second tool exists only to collect a run that outlasted its call.
| Tool | What it does | Arguments | Cost |
|---|---|---|---|
| humanize_texthumanize:write | Rewrite AI-generated text so it reads as human-written while keeping its meaning. Up to 1000 words per call. Costs one credit per word of input. Waits inline for up to five minutes — the longest a run can take — and returns a job id only if a call is cut short. | textultra (Max plan only — a second rewriting pass) | 1 credit / word |
| get_humanize_jobhumanize:read | Read back a run that did not finish inside the humanize_text call. Free — the run was already paid for, and polling costs nothing. | jobId | Free |
// What the assistant sends
{
"name": "humanize_text",
"arguments": {
"text": "The utilization of advanced methodologies facilitates optimal outcomes."
}
}
// What comes back
{
"jobId": "clx8k2p9a0001",
"status": "completed",
"done": true,
"wordCount": 9,
"ultra": false,
"humanizedText": "Using better methods gets you better results.",
"wordsProcessed": 9,
"chunkCount": 1,
"nextStep": "Done. `humanizedText` is the rewritten text — return it to the user as it is."
}Connect it
- Claude (web, desktop, mobile)
- Settings → Connectors → Add custom connector, then paste the endpoint URL. Claude runs the OAuth flow and you sign in with your ReverseGPT account — no key to copy.
- Claude Code
- Run the CLI command below. It registers the server for the current project and handles the OAuth handshake on first use.
- ChatGPT
- Settings → Connectors → Add, then paste the endpoint URL. ChatGPT registers itself and runs the same OAuth flow.
- Cursor, Windsurf, Zed and other MCP clients
- Add the JSON block below to the client's MCP config. These clients generally want a header credential, so use an API key from Settings.
claude mcp add --transport http reversegpt https://www.reversegpt.ai/api/mcp{
"mcpServers": {
"reversegpt": {
"type": "http",
"url": "https://www.reversegpt.ai/api/mcp",
"headers": {
"Authorization": "Bearer rg_live_YOUR_KEY"
}
}
}
}How it behaves
It asks before it spends
A humanize call debits credits, so the tool is advertised as readOnlyHint: false. A client that auto-approves read-only tools will still stop and ask before running this one — which is the point.
It answers in one call
humanize_text holds the call open for up to five minutes — as long as a rewrite can possibly take — and sends progress heartbeats so your client does not time out waiting. So the assistant gets the finished text back in the same call instead of burning a turn on polling. If it does run over, a job id comes back and get_humanize_job collects the result for free.
It is stateless
No session id, nothing retained between requests, and Origin validated on every connection. Up to 1000 words per call — the assistant splits longer drafts and runs the parts.
For implementers
Standard discovery, so a compliant client needs nothing but the server URL.
- MCP endpointhttps://www.reversegpt.ai/api/mcp
- Protected resource metadata (RFC 9728)https://www.reversegpt.ai/.well-known/oauth-protected-resource
- Authorization server metadata (RFC 8414)https://www.reversegpt.ai/.well-known/oauth-authorization-server
- REST API + OpenAPI 3.1https://www.reversegpt.ai/api
Frequently asked questions
- What is the ReverseGPT MCP server?
- It is an implementation of the Model Context Protocol at https://www.reversegpt.ai/api/mcp. Connecting it gives an AI assistant a humanize_text tool, so it can rewrite AI-generated text so it reads as human-written without you leaving the conversation.
- Which AI clients can connect to it?
- Any MCP client that speaks Streamable HTTP — Claude on web, desktop and mobile, Claude Code, ChatGPT, Cursor, Windsurf, Zed, and anything built on the MCP SDKs. Claude and ChatGPT sign in with OAuth; other clients use an API key header.
- How do I authenticate?
- Two ways. Claude and ChatGPT run an OAuth flow, so you just sign in with your ReverseGPT account and never handle a key. Any other client sends an API key from Settings as an Authorization: Bearer header.
- What does it cost to use through an assistant?
- The same as everywhere else: one credit per word of input, drawn from your existing plan. Reading a job back is free. Because a tool call spends credits, the server marks humanize_text as not read-only, so a well-behaved client asks you before it runs.
- Does the assistant see my text?
- Only the text you give it. The tool sends that text to ReverseGPT, gets the rewrite back, and returns it to the conversation — the same path the editor uses. Nothing else in your account is exposed: the server has exactly two tools and neither can read your documents.
- What happens if a rewrite takes a long time?
- humanize_text waits up to five minutes, which is the longest a rewrite can take, and emits progress notifications so a client's own request timeout does not cut it short. That covers every run in practice. If a call is cut short anyway, the run continues on our side and returns a job id with done: false — the assistant polls get_humanize_job until it is ready, and because the run is already paid for, polling costs nothing.
Add the humanizer to your assistant
Create an account, connect the server once, and humanize text without leaving the conversation. Your plan's credits work everywhere.