MCPHub MCP server
2026년 10월 17일 커밋웹 기반 플레이그라운드와 레지스트리를 통해 Model Context Protocol 서버를 탐색, 테스트, 디버깅 및 모니터링할 수 있는 개발자 플랫폼입니다.
다른 이름 namanxdev/MCPHub
- 스타
- 3
- Tools
- 1
- 다운로드, 30일
- —
- 최신 버전
- —
- 마지막 커밋
- 2026년 10월 17일
- 라이선스
- —
MCPHub MCP란?
웹 기반 플레이그라운드와 레지스트리를 통해 Model Context Protocol 서버를 탐색, 테스트, 디버깅 및 모니터링할 수 있는 개발자 플랫폼입니다. 1개 도구를 제공합니다. 예: name.
프로젝트 README에서 생성된 요약
MCPHub MCP server 정보
MCPHub는 AI 에이전트를 위한 Postman과 유사하게 작동하는 Model Context Protocol용 종합 개발자 플랫폼입니다. MCP server를 탐색, 테스트, 디버깅 및 모니터링할 수 있는 중앙 집중식 웹 기반 환경을 제공합니다. 개발자는 대화형 Playground를 사용하여 SSE, streamable HTTP 또는 Desktop Agent를 통한 로컬 stdio server에 연결할 수 있습니다. 플랫폼은 도구 스키마로부터 입력 양식을 자동으로 생성하므로 코드 작성 없이도 테스트가 가능합니다.
단순한 테스트를 넘어, 실시간 JSON-RPC 교환을 확인할 수 있는 Protocol Inspector와 지연 시간 및 에러율을 추적하는 Health Dashboard를 포함합니다. 검색 가능한 Registry를 통해 커뮤니티 서버를 찾을 수 있으며, 실시간 상태 및 보안 등급도 함께 확인할 수 있습니다. 자동화를 위해 CLI 도구를 제공하여 GitHub Actions와 같은 CI/CD 파이프라인 내에서 상태 검사를 실행할 수 있습니다. 또한 원격 서버의 전송 암호화 및 도구 무결성 문제를 감사하는 수동적 보안 스캐너 기능도 갖추고 있습니다.
활용 사례
- 1도구 입력 시각적 테스트개발자는 수동으로 JSON을 구성할 필요 없이 자동 생성된 양식을 사용하여 서버 도구와 상호작용할 수 있습니다.
- 2프로토콜 메시지 디버깅클라이언트와 서버 사이에 흐르는 정확한 JSON-RPC 바이트를 실시간으로 검사할 수 있습니다.
- 3서버 상태 검사 자동화단일 명령어를 사용하여 MCP server 검증을 배포 파이프라인에 직접 통합할 수 있습니다.
- 4서버 보안 감사원격 URL에 대해 수동 스캔을 실행하여 인증되지 않은 도구 노출이나 암호화 문제를 확인할 수 있습니다.
프로젝트의 README를 바탕으로 작성되었습니다. 모든 활용 사례는 해당 내용의 인용문과 대조하여 확인합니다.
MCPHub MCP 도구 1
README에서 감지됨name
| name | — |
구성
AUTH_GITHUB_IDAUTH_GITHUB_SECRETAUTH_GOOGLE_IDAUTH_GOOGLE_SECRETAUTH_SECRETCRON_SECRETDATABASE_URLGITHUB_PERSONAL_ACCESS_TOKEN
| 이름 | 설명 | 유형 |
|---|---|---|
| AUTH_GITHUB_ID | — | |
| AUTH_GITHUB_SECRET | — | |
| AUTH_GOOGLE_ID | — | |
| AUTH_GOOGLE_SECRET | — | |
| AUTH_SECRET | — | |
| CRON_SECRET | — | |
| DATABASE_URL | — | |
| GITHUB_PERSONAL_ACCESS_TOKEN | — |
자동 검사
2026년 11월 14일에 실행됨전체 보고서 →아직 인증되지 않았습니다. 서버는 80점 이상의 점수, 심각한 결함 없음, 그리고 검사에 충분한 데이터를 갖추어야 합니다. 검사 방식 · 점수를 높이기 위해 수정할 항목
- 소스 저장소 연결 및 접근 가능namanxdev/mcphub is public and reachablegithub.com/namanxdev/mcphub ↗통과
- 패키지가 이 저장소를 소스로 명시검사되지 않음
- 빌드 출처(Sigstore 증명)검사되지 않음
- 오픈 소스 라이선스 (OSI)No license결함 발견
- 최근 1년 내 커밋 또는 릴리스Last commit or release 2026-10-17github.com/namanxdev/mcphub ↗통과
- OpenSSF Scorecard검사되지 않음
- 패키지 이름이 인기 패키지와 유사하지 않음검사되지 않음
- 패키지 및 해당 종속성에 중요한 보안 권고 없음검사되지 않음
- 모델에 대한 숨겨진 지침 없음1 tool description and the README read; no override phrases, credential paths or hidden characters통과
- 파괴적인 도구에 라벨 표시됨No tool named like delete, drop or exec통과
- 원격 엔드포인트가 MCP 핸드셰이크에 응답검사되지 않음
README
GitHub에서 보기MCPHub
The developer platform for Model Context Protocol. Discover, test, debug, and monitor any MCP server — from a single browser tab.
mcp-hub-pi.vercel.app · Playground · Registry · Docs
What is MCPHub?
MCP (Model Context Protocol) is the open standard for connecting AI models to external tools and data — think of it as USB-C for AI agents. There are now 1,000+ MCP servers in the wild, built by teams at Anthropic, Microsoft, Stripe, and the broader open-source community.
The problem: the developer tooling hasn't kept up.
Testing an MCP server today means running a local CLI inspector, manually constructing JSON-RPC messages, and getting zero visibility into performance, errors, or reliability. There's no shared registry, no debugging tools, no CI/CD integration.
MCPHub is Postman for MCP. A hosted, web-based platform where you can:
- Connect to any MCP server in seconds
- Test every tool through auto-generated forms — no code required
- Debug protocol-level JSON-RPC exchanges in real time
- Monitor latency, error rates, and uptime over time
- Discover community servers in a live, searchable registry
- Automate health checks in CI/CD with a single CLI command
Features
Playground
Connect to any MCP server (SSE, Streamable HTTP, or stdio via the Desktop Agent) and instantly get an interactive UI for every tool it exposes. MCPHub reads each tool's JSON Schema and renders the right input — text fields, dropdowns, toggles, nested object editors — so you can test without writing a line of code.
- Auto-generated forms from tool
inputSchema - Live response viewer with syntax-highlighted JSON
- Execution history with timestamps and durations
- Save parameter presets for repeated test cases
- Works with any MCP server — no configuration needed
Protocol Inspector
A real-time view of every JSON-RPC message flowing between MCPHub and your server. See the exact bytes on the wire, correlate requests to responses, measure round-trip latency per message.
This is what Chrome DevTools does for HTTP — but for MCP.
- Bidirectional message stream (→ requests, ← responses)
- Syntax-highlighted JSON with collapsible trees
- Latency timing per message
- Filter by method, direction, or keyword
- Pause, resume, and clear the stream
Health Dashboard
Per-tool performance metrics collected automatically during every session and from scheduled health checks. Know at a glance whether a server is fast, reliable, and still online.
- P50 / P95 / P99 latency per tool
- Error rate and error type breakdown
- Response size distribution
- Uptime history
- Embeddable status badges for README files
Registry
A searchable, community-driven directory of MCP servers — with live health data attached to every entry. Not a static markdown file. Every server in the registry is periodically health-checked and gets a status badge reflecting its current state.
- Full-text search by name, description, or capability
- Filter by transport type, category, and health status
- "Test in Playground" — click to connect directly
- Community submissions via GitHub OAuth
- Status badges you can embed in your own docs
- Security grade (A–F) from the latest scan, shown on every server's detail page
CLI Tool
Run a full MCP health check from the terminal or CI/CD pipeline:
npx mcphub test https://your-server.example.com/sse
MCPHub Test — https://your-server.example.com/sse
Transport: sse | 2026-06-22T10:30:00Z
✓ Connection (142ms) Connected to my-server v1.2.0
✓ Initialize Handshake (130ms) Protocol: 2024-11-05
✓ Protocol Version (0ms) Supported
✓ Tools List (85ms) 5 tools found
✓ Tool Schemas (2ms) All 5 schemas valid
✓ Resources List (62ms) 2 resources found
✓ Prompts List (58ms) 1 prompt found
─────────────────────────────────────────
7/7 checks passed — 479ms total
Capabilities: 5 tools · 2 resources · 1 prompt
Exit code 0 on pass, 1 on any failure. Plug it straight into GitHub Actions:
- name: Test MCP server
run: npx mcphub test http://localhost:3001/sse --junit results.xml
Flags:
| Flag | What it does |
|---|---|
--json |
Machine-readable JSON output |
--smoke-test |
Invoke each tool and verify it doesn't crash the server |
--verbose |
Print raw JSON-RPC messages |
--timeout <ms> |
Per-check timeout (default: 30 000) |
-H "Key: Value" |
Custom request headers (repeatable) |
--junit <path> |
JUnit XML output for CI test reporters |
--watch |
Re-run on interval during development |
Security Scan
A passive, read-only auditor for remote MCP servers. Point it at any Streamable-HTTP server — an ad-hoc URL or one already in the registry — and it inspects the common auth, transport, and tool-integrity holes, then emits a scored JSON + Markdown report. The same engine can scan the whole registry and rank servers into a "State of MCP Security" leaderboard.
# Scan any server by URL (no database required)
npm run scan -- scan https://your-server.example.com/mcp --no-persist
# Seed the registry from the official MCP registry, then scan a registered server
npm run registry:seed
npm run scan -- scan --registry <server-id>
# Scan every registered streamable-http server and write a ranked leaderboard
npm run scan -- scan --batch --md SCAN-RESULTS.md
What it checks
| # | Check | Detects |
|---|---|---|
| 1 | Transport encryption | Plaintext http:// vs TLS |
| 2 | Unauthenticated tool exposure | Tri-state: exposed (serves ≥1 tool with no credentials — critical) / enforced (handshake rejected — pass) / inconclusive (handshake accepted but 0 tools — advisory, never critical) |
| 3 | OAuth 2.1 metadata | /.well-known/oauth-protected-resource and /.well-known/oauth-authorization-server discovery |
| 4 | RFC 8707 audience binding | Resource indicators in the OAuth metadata (mitigates confused-deputy token replay) |
| 5 | Tool-description poisoning | Hidden model-directed instructions, prerequisite-action injection, instruction overrides, sensitive-file/cred references, exfiltration directives, and invisible/bidi/tag Unicode — tuned for low false positives |
| 6 | Rug-pull / silent redefinition | SHA-256 fingerprint of each tool's name + description + inputSchema, baselined per target and diffed on re-scan; any post-baseline mutation is flagged |
| 7 | Advisory checklist | Scope minimization, short-lived tokens, no token passthrough, server-side audit logging — surfaced as advisory (cannot be verified externally) |
Scoring: start at 100, subtract per fail by severity (critical 40 / high 20 / medium 10 / low 4), floored at 0 → A ≥90 · B ≥75 · C ≥60 · D ≥40 · F <40.
Passive / ethics guarantee. The scanner only sends initialize, notifications/initialized, tools/list, and GETs the .well-known documents. It never executes a tool, never sends credentials, and never scans private/internal hosts — an outbound-URL guard rejects localhost and private IP ranges before any request. Only scan servers you are authorized to test. stdio transport is a documented stub for a later pass; this release audits Streamable HTTP.
The
SCAN-RESULTS.mdleaderboard is generated by runningregistry:seedthenscan --batchagainst a configuredDATABASE_URL.
In the registry UI. Every persisted scan is surfaced on the server's registry detail page: a Security card sits alongside the health metrics, showing the latest grade, score, and a one-line auth summary (e.g. "13 tools exposed, no auth" or "auth enforced"). Servers that haven't been scanned yet simply show —.
Desktop Agent
The deployed MCPHub app runs in the cloud, but your local MCP servers don't. The Desktop Agent is a lightweight bridge — install it once, and the web app can reach localhost just like Postman's desktop agent.
npm install -g @naman_411/mcphub-agent
mcphub-agent start
When the agent is running, the Playground shows a ⚡ DESKTOP AGENT DETECTED banner. Enable it and connect to any local server — via URL or stdio command — directly from the browser.
Why MCPHub?
Existing tools each solve one piece of the problem:
| MCP Inspector | Smithery.ai | awesome-mcp-servers | Postman | MCPHub | |
|---|---|---|---|---|---|
| Server discovery / registry | — | ✓ | ✓ | — | ✓ |
| Connect to arbitrary servers | ✓ | — | — | — | ✓ |
| Auto-generated tool forms | ✓ | — | — | — | ✓ |
| Protocol message inspection | — | — | — | — | ✓ |
| Health & latency monitoring | — | — | — | — | ✓ |
| Web-based (shareable) | — | ✓ | ✓ | ✓ | ✓ |
| CI/CD automation | — | — | — | ✓ | ✓ |
| Open source / self-hostable | ✓ | — | ✓ | — | ✓ |
MCPHub is the only tool that does all of these together.
Getting Started
Use the hosted version
Open mcp-hub-pi.vercel.app — no account required for the Playground or Registry.
Run locally
1. Clone and install
git clone https://github.com/yourusername/mcphub
cd mcphub
npm install
2. Configure environment
cp .env.example .env
# Database (Neon PostgreSQL)
DATABASE_URL=postgresql://user:password@ep-xxx.region.neon.tech/dbname?sslmode=require
# Cron job protection
CRON_SECRET=your-secret-here
# GitHub PAT (for registry GitHub integration)
GITHUB_PERSONAL_ACCESS_TOKEN=github_pat_...
# NextAuth.js
AUTH_SECRET=generate-with-openssl-rand-base64-32
AUTH_GITHUB_ID=your-github-oauth-client-id
AUTH_GITHUB_SECRET=your-github-oauth-client-secret
AUTH_GOOGLE_ID=your-google-oauth-client-id
AUTH_GOOGLE_SECRET=your-google-oauth-client-secret
3. Set up the database
npm run db:setup # Create tables
npm run db:seed # Seed the registry
4. Start
npm run dev
Open http://localhost:3000.
npm run dev automatically uses a shared Local Developer session. No sign-in,
OAuth credentials, or auth database records are needed for local authentication.
Registry and metrics features still require DATABASE_URL. The bypass is enabled
only when NODE_ENV=development; production builds use normal authentication.
If Google sign-in reports invalid_client / "The provided client secret is
invalid", replace AUTH_GOOGLE_SECRET with the secret belonging to the OAuth
client identified by AUTH_GOOGLE_ID, then restart or redeploy the app.
OAuth Setup
GitHub
- github.com/settings/developers → OAuth Apps → New OAuth App
- Callback URL:
http://localhost:3000/api/auth/callback/github(or your Vercel domain) - Copy Client ID →
AUTH_GITHUB_ID, Client Secret →AUTH_GITHUB_SECRET
- console.cloud.google.com → APIs & Services → Credentials → Create OAuth Client ID
- Application type: Web application
- Redirect URI:
http://localhost:3000/api/auth/callback/google - Copy Client ID →
AUTH_GOOGLE_ID, Client Secret →AUTH_GOOGLE_SECRET
Scripts
npm run dev # Start dev server
npm run build # Production build
npm run lint # ESLint
npm run test # Vitest
npm run db:setup # Create database tables
npm run db:push # Push Drizzle schema changes
npm run db:seed # Seed registry data
npm run scan # Security Scan CLI (scan <url> | --registry <id> | --batch)
npm run registry:seed # Import servers from the official MCP registry
Deploy to Vercel
- Push to GitHub and import the repo at vercel.com
- Add all
.envvariables in Settings → Environment Variables - Redeploy
The included vercel.json registers a cron job that runs /api/cron/health-check daily at midnight UTC to refresh registry health data.
Tech Stack
| Framework | Next.js 16 (App Router, React 19) |
| Language | TypeScript 5 |
| Styling | Tailwind CSS 4 + shadcn/ui |
| State | Zustand 5 |
| Database | Neon PostgreSQL + Drizzle ORM |
| Auth | NextAuth.js v5 |
| MCP | @modelcontextprotocol/sdk |
| Animations | Framer Motion |
| Charts | Recharts |
| Syntax highlighting | Shiki |
| Deploy | Vercel |
Project Structure
app/ # Next.js App Router
├── page.tsx # Landing page
├── playground/ # Tool playground
├── inspector/ # Protocol inspector
├── dashboard/ # Health dashboard
├── registry/ # Server registry
├── docs/ # Documentation
└── api/ # API routes (MCP proxy, registry, auth, cron)
components/ # React components (playground, inspector, registry, etc.)
lib/ # Server logic (ConnectionManager, ProtocolLogger, Drizzle)
├── security/ # Security Scan engine (probe, checks, poisoning, scoring)
stores/ # Zustand state (connection, playground, inspector)
hooks/ # Custom React hooks
cli/ # CLI package — `npx mcphub test`
agent/ # Desktop Agent — `mcphub-agent`
scripts/ # DB setup, seeding, and the Security Scan CLI
Desktop Agent — Full Docs
See DESKTOP_AGENT_QUICKSTART.md for installation, usage, and troubleshooting.
License
MIT
MCPHub MCP 질문
MCPHub MCP는 어떤 도구를 제공하나요?
1개 도구를 제공합니다. 예: name.
MCPHub란 무엇인가요?
MCPHub는 모든 MCP server에 연결, 테스트, 디버깅 및 모니터링할 수 있도록 설계된 호스팅 웹 기반 플랫폼입니다.
로컬 MCP server도 테스트할 수 있나요?
네, 웹 앱과 localhost 사이의 브릿지 역할을 하는 Desktop Agent를 사용하여 로컬 서버에 연결할 수 있습니다.
보안 스캔은 어떻게 작동하나요?
보안 스캔은 도구를 실행하지 않고 전송 암호화, 인증되지 않은 도구 노출 및 기타 취약점을 검사하는 수동적 읽기 전용 감사 도구입니다.
플레이그라운드에서 도구를 테스트하려면 코드를 작성해야 하나요?
아니요, 플레이그라운드는 도구의 JSON Schema를 기반으로 대화형 양식을 자동으로 생성합니다.