Skip to main content
Neural Defend

NeuroVerify Developer Quickstart

Obtain a NeuroVerify API key through Neural Defend onboarding, store it outside source control, and choose the official Python SDK, TypeScript SDK, MCP server, or direct REST API. Test with media you are authorized to upload. For every response, branch on status before reading risk fields, retain the transaction identifier for correlation, and send medium or high risk to an appropriate review process. Do not treat a rejection as an authenticity result.

Maintained byNeural Defend
Updated August 7, 2026
Siva Shankar, Co-founder and CTO of Neural Defend.
Siva ShankarCo-founder & CTO
An article-specific diagram for NeuroVerify Developer Quickstart organizes key sections including confirm prerequisites, send one bounded request, verify behavior before production.

Confirm prerequisites

Request access through the Neural Defend website or contact support for an existing account. Put the issued key in a secret manager or the documented environment variable for your client. Never paste a real key into source files, browser bundles, prompts, tool arguments, screenshots, or support tickets.

Choose a representative test file only after confirming that your organization is authorized to send it for analysis. Media may contain biometric or other sensitive personal data, so the upload must fit your own privacy, retention, logging, and access-control obligations.

  • Python: install the pinned neuraldefend release in the application environment.
  • Node.js: install @neuraldefend/sdk and keep the key server-side.
  • MCP: install neuraldefend-mcp and configure a narrow allowed-directory list.
  • REST: use the production or staging server listed in the OpenAPI contract.

Send one bounded request

For an image, use the image method in an SDK or POST multipart form data to /detect/image with one file. For a video, use the video method or POST to /detect/video. The TypeScript and Python clients stream or prepare uploads according to their runtime-specific input rules, which is safer than duplicating low-level transport logic.

Allow the HTTP client to construct the multipart boundary. For direct REST calls, send the API key in x-api-key and parse the named response envelope for the selected media type.

cURL
curl -X POST "YOUR_API_ORIGIN/detect/image" \
  -H "x-api-key: YOUR_API_KEY" \
  -F "file=@selfie.jpg"
Direct REST call. Replace YOUR_API_ORIGIN and YOUR_API_KEY with the API address and key your team receives at onboarding. curl builds the multipart body.
Python
# pip install neuraldefend==1.0.2
# The client reads your key from the NEURALDEFEND_API_KEY environment variable.
from neuraldefend import NeuroVerifyClient

with NeuroVerifyClient() as client:
    result = client.detect_image("selfie.jpg")

if result.scored:
    print(result.to_dict())
elif result.rejected:
    print("Not scored. Ask for a new image.")
else:
    print("No decision. Hold the case.")
Official Python SDK, run server-side. Branch on scored and rejected before reading any risk field.
TypeScript
// npm install @neuraldefend/sdk@1.0.3
// Node.js reads your key from the NEURALDEFEND_API_KEY environment variable.
import { NeuroVerifyClient } from "@neuraldefend/sdk";

const client = new NeuroVerifyClient();
const result = await client.detectImage("selfie.jpg");

if (result.status === "success") {
  console.log(JSON.stringify(result));
} else if (result.status === "rejected") {
  console.log("Not scored. Ask for a new image.");
} else {
  console.log("No decision. Hold the case.");
}
Official TypeScript SDK on Node.js 22 or newer. Keep the key on your server; never ship it in browser code.
An article-specific topic map for NeuroVerify Developer Quickstart organizes key sections including send one bounded request, verify behavior before production, confirm prerequisites.

Verify behavior before production

A successful score, a business rejection, and a transport or service failure are different outcomes. Confirm that your code does not read nullable score fields before checking status. Also confirm that a low-risk result does not bypass other controls and that medium, high, or unknown results have a review owner.

The API is not idempotent. The official clients apply narrow retry policies for selected server and rate-limit responses. Adding a second automatic retry layer can create extra transactions, so account for the client's behavior before wrapping it.

Sample response
{
  "unified_face_authenticity_score": {
    "unique_trx_id": "…",
    "filename": "selfie.jpg",
    "content_type": "image/jpeg",
    "status": "success",
    "status_code": …,
    "billable": …,
    "risk_score": …,
    "risk_level": "medium",
    "message": "…"
  }
}
Shape of a scored image response. Values shown as … vary by request. risk_score runs from 0.1 to 10.0, risk_level is low, medium or high, and scored responses may also include ai_threat_signals.
Sample response
{
  "unified_face_authenticity_score": {
    "unique_trx_id": "…",
    "filename": "selfie.jpg",
    "content_type": "image/jpeg",
    "status": "rejected",
    "status_code": 6,
    "billable": true,
    "risk_score": null,
    "risk_level": null,
    "message": "…"
  }
}
A no-face rejection arrives on HTTP 200 with status_code 6. It is billable and carries no score, so null never means low risk. Multiple faces use status_code 7.

Evidence boundary

Limitations

  • This quickstart does not include a live credential, tenant-specific limits, a browser credential-issuance design, or approval to upload any particular file.

Source record

Sources and references

  1. Neural DefendNeuralDefend SDKs quick startVerified August 7, 2026.
  2. Neural DefendNeuroVerify Detection API OpenAPI contractVerified August 7, 2026.

Choose the contract

Review the API before implementation

Confirm the endpoint, envelope, media rules, and retry behavior for your first request.

Read the API overview