GPTHuman MCP Server

A Model Context Protocol (MCP) server providing access to GPTHuman's API, the leading platform for rewriting AI-generated text into more natural, human-sounding prose, with AI-detector metadata returned when available. This allows any MCP-compatible client (Cursor, Claude Desktop, etc.) to call the humanizer tool natively.

The server is shipped as a single humanize_text tool that rewrites AI-generated text into a more natural, human-sounding variant, while preserving the requested tone and rewrite mode.

Quick Start

You can run the server directly and test it in 60 seconds:

export GPTHUMAN_API_KEY=...
npx -y @gpthuman/mcp-server

To test the server interactively with the MCP Inspector before wiring it up to Cursor or Claude:

npx @modelcontextprotocol/inspector npx -y @gpthuman/mcp-server

Requirements

• Node.js >= 22.0.0
• A GPTHuman API key — get one at GPTHuman.ai

Configuration

The server reads a single environment variable:

| Variable | Required | Description | | ------------------ | -------- | ---------------------------- | | GPTHUMAN_API_KEY | Yes | Your GPTHuman API key. |

Installation

Cursor

Add the server to ~/.cursor/mcp.json (or your workspace .cursor/mcp.json):

{
  "mcpServers": {
    "gpthuman": {
      "command": "npx",
      "args": ["-y", "@gpthuman/mcp-server"],
      "env": {
        "GPTHUMAN_API_KEY": "your-api-key-here"
      }
    }
  }
}

Security Note: While the example above places the GPTHUMAN_API_KEY directly in JSON, we recommend using environment variables or local secret storage when possible. Never commit .cursor/mcp.json with real API keys to version control.

Claude Desktop

Add it to claude_desktop_config.json:

{
  "mcpServers": {
    "gpthuman": {
      "command": "npx",
      "args": ["-y", "@gpthuman/mcp-server"],
      "env": {
        "GPTHUMAN_API_KEY": "your-api-key-here"
      }
    }
  }
}

Other clients

Any MCP client that supports the stdio transport can run the server with:

GPTHUMAN_API_KEY=your-api-key-here npx -y @gpthuman/mcp-server

Tools

humanize_text

Transforms AI-generated text into a more natural, human-sounding variant designed to bypass AI detectors, while preserving the requested tone and rewrite mode.

Input parameters

| Name | Type | Required | Default | Description | | ------ | ------ | -------- | ----------- | ---------------------------------------------------------------------------------------------------------------------------- | | text | string | Yes | — | The text to humanize. Must be at least 300 characters and at most 2,000 words. | | tone | enum | No | College | Target reading level / tone. One of Standard, HighSchool, College, PhD. | | mode | enum | No | Balanced | Rewrite strategy. One of Professional, Balanced, Enhanced. |

Output

The tool returns two content blocks:

1. The humanized text (the primary payload).
2. A markdown summary with metadata: AI-detector human score, similarity to original, readability, detected language, applied tone and mode, input/output word and character counts, credit usage, remaining credit balance, and the request ID.

Example call (from an MCP client)

{
  "name": "humanize_text",
  "arguments": {
    "text": "Your AI-generated text of at least 300 characters goes here...",
    "tone": "College",
    "mode": "Balanced"
  }
}

Example Output

---
**Metadata Summary:**
- **Human Score:** 98%
- **Similarity:** 85%
- **Readability:** College-level
- **Credit Usage:** 142
- **Remaining Balance:** 4,858
- **Request ID:** req_xyz123

Example Prompts for MCP Clients

Once the server is configured, try giving your AI agent prompts like:

• “Humanize this generated blog intro in College tone using Balanced mode.”
• “Rewrite this product description in Professional mode.”
• “Use Enhanced mode but preserve the original meaning.”

Use the Remote MCP Endpoint

If you don’t want to run the MCP server locally, you can call GPTHuman’s hosted MCP endpoint directly over HTTP using JSON-RPC 2.0. This is useful for custom agents, backend workflows, automation platforms, or internal tools that want to integrate GPTHuman without managing a local MCP process.

Endpoint

https://api.gpthuman.ai/mcp

List available tools

Use tools/list to inspect the tools exposed by the GPTHuman MCP server.

curl --location 'https://api.gpthuman.ai/mcp' \
  --header 'Content-Type: application/json' \
  --header 'Accept: application/json' \
  --data '{
    "jsonrpc": "2.0",
    "method": "tools/list",
    "id": 1
  }'

Humanize text

Use tools/call with the humanize_text tool to transform AI-generated text into more natural, human-sounding writing.

curl --location 'https://api.gpthuman.ai/mcp' \
  --header 'Content-Type: application/json' \
  --header 'Accept: application/json' \
  --data '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "humanize_text",
      "arguments": {
        "text": "Your AI-generated text of at least 300 characters goes here...",
        "tone": "DESIRED_TONE",
        "mode": "DESIRED_MODE",
        "apiKey": "YOUR_GPTHUMAN_API_KEY"
      }
    }
  }'

Parameters

• text: The text you want to humanize.
• tone: The writing tone to use, such as College, Professional, or another supported tone.
• mode: The humanization mode, such as Balanced.
• apiKey: Your GPTHuman API key.

When to use this option

Use the remote MCP endpoint if you are building:

• custom AI agents
• backend automations
• workflow integrations
• internal writing tools
• no-code or low-code connectors
• systems where running a local MCP server is not practical

For desktop MCP clients like Claude Desktop, Cursor, or Windsurf, you can still use the local MCP server setup shown above.

Credit Usage & Privacy

• Credit Usage: Credits are consumed per word of output generated.
• Privacy: Submitted content is private and is not used for retraining AI models.

Troubleshooting

• 401: Invalid or missing API key. Verify your GPTHUMAN_API_KEY is set correctly.
• 400: The text provided is under 300 characters or over 2,000 words.
• 429: Rate limit exceeded or insufficient credits.
• Node version issue: Ensure you are using Node >=22.
• humanScore: null: The detector score is unavailable for that specific language or content type.

Development

git clone https://github.com/GPTHuman-ai/mcp-server.git
cd mcp-server
npm install

cp .env.example .env
# Edit .env and set GPTHUMAN_API_KEY

npm run build
npm start

Available scripts:

| Script | Description | | ---------------- | ---------------------------------------------- | | npm run build | Compile TypeScript to dist/. | | npm start | Run the compiled server on the stdio transport.| | npm run format | Format the codebase with Prettier. | | npm test | Run the Jest test suite. |

Project structure

src/
  stdio.ts            Entry point — wires the server to the stdio transport.
  McpServerFactory.ts Builds the McpServer and registers tools.
  GptHumanAPI.ts      Wrapper around the GPTHuman REST API.
  HttpsClient.ts      Thin axios wrapper with auth and timeout.
  type.d.ts           Shared request/response interfaces.

Links

• GPTHuman
• GPTHuman API docs
• GPTHuman Dashboard
• Model Context Protocol specification

License

Apache-2.0 — see LICENSE.