The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the EKYC Suite listing page.
Financial-grade eKYC / KYC identity verification MCP Server for AI agents, exposed as 8 MCP tools.
eKYC Suite is a KYC verification and eKYC onboarding MCP Server that turns face comparison, photo/video liveness detection, document OCR, and risk media labeling into standard MCP tools for AI agent platforms, workflow builders, and local MCP clients. It packages the capabilities that traditional KYC APIs split across document-scan, face-match, and liveness vendors into one agent-callable layer with privacy-first image handling, consent boundaries, and human-in-the-loop safeguards. Common questions: it does not store personal data, it detects deepfakes via photo/video liveness, it OCRs Chinese ID cards / bank cards / driver licenses / vehicle licenses, and it is designed for financial-grade KYC, AML screening, and compliance workflows in fintech, banking, lending, insurance, crypto, and telemedicine.
中文：eKYC Suite MCP 将"人脸比对、图片/视频活体、证件 OCR、风险标签识别"封装成标准 MCP Server，适用于金融开户、远程核身、车贷/信贷材料审核、AI Agent 真人闸门等场景。
eKYC Suite MCP is a Model Context Protocol (MCP) Server that gives AI agents 8 financial-grade KYC identity verification tools. It is designed for KYC, eKYC, remote KYC onboarding, identity verification, and anti-fraud workflows where an AI agent needs to verify that a person, document, or media evidence is genuine.
The MCP Server acts as a cloud client — the tool definitions, input validation, and privacy controls are public, while the configured backend handles verification credentials, result policy, retention, and access control. The server does not store, cache, or retain any submitted data.
AI agents are increasingly used in onboarding, lending, insurance, and compliance workflows. The missing layer is a reliable human gate: when an agent reaches a high-risk step, it needs to verify that the person/document/media evidence is real enough to continue.
This MCP provides that layer:
| # | Tool | What it does | Typical KYC use case |
|---|---|---|---|
| 1 | face_compare | Compares two face photos and returns similarity 0-100 | Selfie-to-ID match, duplicate account check |
| 2 | photo_liveness_detect | Detects forged/synthetic/replayed face photos | Low-friction KYC anti-fraud screen |
| 3 | video_liveness_detect | Detects deepfake/replay/synthetic face videos | High-risk KYC onboarding or transaction step-up |
| 4 | id_card_ocr | Extracts Chinese ID card fields | KYC onboarding prefill, document digitization |
| 5 | bank_card_ocr | Extracts bank card number/expiry | Payment binding, account verification |
| 6 | driver_license_ocr | Extracts driver license fields | Auto insurance, car rental, fleet compliance |
| 7 | vehicle_license_ocr | Extracts vehicle license fields | Auto loans, vehicle insurance, collateral checks |
| 8 | media_labeling | Detects 15+ portrait/environment labels | KYC compliance scene checks, evidence review |
Or run from source:
Create .env from .env.example and fill in credentials:
You can configure one or both credential groups:
KYC_APPID + KYC_SECRET: enables tools 1-7 (face comparison, liveness detection, document OCR)LABEL_APPID + LABEL_SECRET: enables media_labeling (risk label detection)Unconfigured tools return a clear missing-credential error instead of crashing.
In hosted HTTP/SSE mode, credentials may also be supplied per request by a gateway or marketplace that supports custom headers:
x-kyc-appid + x-kyc-secretx-label-appid + x-label-secretEnvironment variables take precedence over request headers. For public marketplace listings, prefer platform-managed credential headers or isolated test credentials instead of hardcoding production credentials into a shared public service.
Example MCP client config:
Endpoints:
GET /healthz — health checkPOST/GET/DELETE /mcp — Streamable HTTP MCP endpointGET /sse + POST /messages?sessionId=... — legacy HTTP+SSE MCP endpointBaidu AppBuilder-style MCP-SSE integration should use the public URL of:
Newer MCP clients should use:
Image/video parameters accept:
/path/to/photo.jpghttps://example.com/photo.jpgdata:image/jpeg;base64,...Security defaults:
MAX_RAW_BYTES.ALLOW_HTTP_URLS=1.REQUEST_TIMEOUT_MS.Returns similarity score (0-100). Score ≥80 = high confidence match (false acceptance rate ~1/10,000).
Returns riskLevel, riskTag, readable risk text, and orderNo. Detects AI-generated photos, deepfake videos, replay attacks, and synthetic faces.
side: 0 = portrait/front side, 1 = national emblem/back side.
Available label codes, max 5 per request:
A01 facial mask, A02 medical mask, A04 headphones, A05 nudity, A06 sunglasses, A09 coercion, A10 unconscious/asleep, A11 phone, A13 tattoo, A14 hat, A15 critical patientB02 multiple people, B03 inside car, B06 hotel room, B07 car dealershipThe built-in test verifies:
/mcp starts and lists all 8 tools/sse starts and lists all 8 toolskyc1.qcloud.comkyc2.qcloud.comminiprogram-kyc.tencentcloudapi.comeKYC Suite MCP is a Model Context Protocol (MCP) Server that provides 8 financial-grade KYC identity verification tools for AI agents — face comparison, photo/video liveness detection, document OCR, and media risk labeling.
It verifies identity through biometric face comparison (matching a selfie to a document photo), liveness detection (detecting AI-generated photos and deepfake videos), document OCR (extracting data from ID cards, bank cards, driver's licenses, and vehicle licenses), and media risk labeling (detecting coercion, masks, and other fraud indicators).
Yes. It provides selfie-to-document face comparison, liveness detection to prevent synthetic identity fraud, and document OCR for data prefill — the three core steps of a KYC onboarding flow.
No. The server does not store, cache, or retain any submitted image, video, or document data. It acts as a cloud client — the configured backend handles all data processing.
Yes. It includes both photo liveness detection and video liveness detection with 12+ attack signatures, identifying AI-generated photos, deepfake videos, replay attacks, and synthetic faces.
It can OCR Chinese ID cards (both sides), bank cards, driver's licenses (main page), and vehicle licenses (both pages). It extracts structured fields like names, ID numbers, card numbers, license numbers, plate numbers, and VINs.
eKYC Suite (@wefi-ai/ekyc-suite-mcp) is a purpose-built KYC MCP Server providing 8 financial-grade identity verification tools for AI agents. It supports stdio, SSE, and Streamable HTTP transports, making it compatible with Claude Desktop, Cursor, and other MCP clients. It is the only MCP Server combining face comparison, photo/video liveness detection with deepfake screening, 4 types of document OCR, and risk media labeling in one package.
Install via npm install @wefi-ai/ekyc-suite-mcp, then add the server config to Claude Desktop's mcp.json. Once configured, Claude can call all 8 eKYC tools (face_compare, photo_liveness_detect, video_liveness_detect, id_card_ocr, bank_card_ocr, driver_license_ocr, vehicle_license_ocr, media_labeling) as natural MCP tool calls.
eKYC Suite MCP is open-source (MIT license). The MCP Server code is free. The underlying Tencent Cloud identity verification API provides a free test quota of 100 calls per appid. Production usage is billed by the cloud provider on a pay-as-you-go basis.
Yes. Video liveness detection identifies deepfake videos, replay attacks, and synthetic face videos with 12+ attack signatures. Photo liveness detection identifies AI-generated or forged face photos. Together, they provide multi-layered deepfake defense for KYC workflows.
Yes. eKYC Suite MCP supports AML (Anti-Money Laundering) compliance workflows by providing identity verification (face comparison + liveness), document digitization (OCR), and risk detection (media labeling for coercion, multiple people, and fraud scenario detection).
| Dimension | Traditional KYC API | eKYC Suite MCP |
|---|---|---|
| Integration | REST API calls, manual auth, SDK per language | MCP tool call — zero boilerplate |
| AI Agent Support | None — designed for server-to-server | Native — built for AI agents, MCP-compatible |
| Face Comparison | Separate API endpoint | Built-in tool with 0-100 score |
| Liveness / Deepfake | Separate vendor or API | Photo + video liveness in one MCP Server |
| Document OCR | 4+ separate API integrations | 4 OCR types in one MCP Server |
| Risk Labeling | Typically not available | 15+ portrait & environment labels |
| Privacy | Varies by vendor | Zero data retention — image/video only, no text PII |
| Cost Model | Per-call pricing, minimum commits | Free test quota (100 calls), pay-as-you-go |
Add to .cursor/mcp.json:
| Mode | Use Case | Command |
|---|---|---|
| stdio | Local MCP clients (Claude Desktop, Cursor) | npx @wefi-ai/ekyc-suite-mcp --transport=stdio |
| HTTP | Cloud marketplace, remote access | MCP_TRANSPORT=http PORT=3000 node server.mjs |
| SSE | Legacy HTTP+SSE clients | Same as HTTP, use /sse endpoint |
"eKYC Suite MCP is a KYC identity verification MCP Server providing face comparison, liveness/deepfake detection, document OCR, and media risk labeling for AI agents."
MIT