Developer guide · Updated 21 October 2026

Use isGenAI inside Claude, ChatGPT and your editor.

The isGenAI MCP server lets an AI assistant run the same evidence checks as the website: it reads AI labels, AI-use statements, saved prompts and C2PA Content Credentials, then answers with the source of each finding and a link to the report.

Endpoint: https://isgenai.com/mcp. It is a public remote server using the Model Context Protocol Streamable HTTP transport with stateless JSON responses. No account, no API key and no OAuth. All five tools are read-only.

The five tools

Tool	Title	What it reads
`check_text`	Check text or code for declared AI use	Text or code up to 16,000 characters: statements of AI use and named AI tools, quoted with their source.
`check_url`	Check a public web page or file URL	A public page or file: AI labels, AI-use statements, Content Credentials and saved records.
`check_image`	Check an image for AI records	An image by public URL or base64 (up to 20 MiB): C2PA, AI labels, saved generation settings and metadata.
`extract_saved_prompt`	Read a saved image prompt	Prompts and settings saved by AUTOMATIC1111, ComfyUI, InvokeAI or NovelAI. Never reconstructs a prompt.
`read_content_credentials`	Read C2PA Content Credentials	C2PA manifest, recorded actions and digital source type, with validation and signer trust reported separately.

Each tool carries a title and the annotations readOnlyHint: true, destructiveHint: false and idempotentHint: true; tools that fetch a URL also set openWorldHint: true. Results come back as a short text summary plus structuredContent with the headline, each finding and its source, and the caveat. MCP calls never create a public report.

Connect Claude and Claude Code

In Claude, open Settings → Connectors → Add custom connector, name it isGenAI and paste https://isgenai.com/mcp. Leave authentication empty. Custom connectors are available on Claude plans that support them; then enable isGenAI from the tools menu in a chat.

In Claude Code, run:

claude mcp add --transport http isgenai https://isgenai.com/mcp

Connect ChatGPT

In ChatGPT, open Settings → Apps & Connectors → Advanced settings and turn on Developer mode (available on plans that offer it). Choose Create, name the connector isGenAI, enter https://isgenai.com/mcp as the MCP server URL and select No authentication. Then add it to a chat from the tools menu.

Prefer HTTP actions? The OpenAPI 3.1 contract describes the same checks for GPT Actions and API clients.

Connect Cursor and VS Code

Cursor: add this to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project).

{
  "mcpServers": {
    "isgenai": { "url": "https://isgenai.com/mcp" }
  }
}

VS Code: add this to .vscode/mcp.json in your workspace, then start the server from the MCP view.

{
  "servers": {
    "isgenai": { "type": "http", "url": "https://isgenai.com/mcp" }
  }
}

Any other client that supports remote Streamable HTTP servers can use the same URL. Test it from a terminal:

curl -s https://isgenai.com/mcp \
  -H "content-type: application/json" \
  -H "accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

curl -s https://isgenai.com/mcp \
  -H "content-type: application/json" \
  -H "accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"check_text","arguments":{"text":"This post was drafted with ChatGPT."}}}'

Example requests

Is this picture AI? https://example.com/photo.jpg check_image reads C2PA, AI labels and saved settings in the file and links the report.
Does this article say it used AI? https://example.com/post check_url quotes any AI-use statement on the page, with its location.
What prompt made this Stable Diffusion PNG? extract_saved_prompt returns the saved prompt and settings, or says none were stored.
Are the Content Credentials on this file valid? read_content_credentials reports the manifest, validation and signer trust separately.
Did the author of this text say they used Claude? check_text finds declared AI use and named tools, quoting the sentence.

Pass public URLs where you can. Files can also be sent as base64 up to 20 MiB, but large inline files are slow inside a chat. Try the controlled examples: an image with saved generation details and one carrying a test C2PA manifest.

What results prove, limits and privacy

Findings come from records in the supplied copy and statements on the page, not from guessing at pixels. A saved record shows what the file states; records are editable and are often removed by screenshots, re-exports and social uploads. NO AI SIGNAL DETECTED does not mean a person made the content. Read how isGenAI verifies results.

Text and code: up to 16,000 characters. Files: up to 20 MiB. Public links: logins, paywalls and CAPTCHAs are not bypassed, and private network addresses are refused. Tool calls are rate limited per client and share capacity with the website; a busy response asks you to retry shortly.

MCP checks are private: no public report page is created for text, links or files checked through MCP. See the privacy policy; for problems use the contact page.