vantagate-mcp-server

vantagate-mcp-server

Human-in-the-Loop authorization gateway for AI Agents. Securely pause MCP workflows and route high-risk actions to human approvers via Slack or Email.

Category
访问服务器

README

VantaGate MCP Server & OpenAPI Integration

npm version License: MIT Node.js >= 20

Human-in-the-Loop authorization for AI Agents. VantaGate intercepts high-risk actions, routes them to a human approver via Slack or Email, and returns a cryptographically-signed decision - so your agent resumes or halts with a full audit trail.


What is VantaGate?

AI agents are increasingly capable of executing consequential real-world actions: sending emails to thousands of users, deleting database records, moving money, deploying to production. VantaGate is the trust layer that ensures humans remain in control.

Agent hits risky action
        ↓
POST /v1/checkpoint  ──→  Human receives Slack/Email notification
        ↓                          ↓
   id + dashboard_url    Human reviews payload & decides
        ↓                          ↓
Poll GET /v1/checkpoint/{id}  ←──  Decision recorded + payload purged
        ↓
   APPROVED → Resume workflow
   REJECTED → Halt + report reason

Key Guarantees

  • < 100ms API response - your agent is never blocked waiting for I/O
  • AES-256 payload encryption at rest - payload is permanently destroyed after the decision
  • Zero-Retention polling - GET /checkpoint/{id} never returns the original payload
  • Cryptographic audit trail - every decision is HMAC-SHA256 signed
  • Stateless protocol - no SDK required; plain HTTP from any language

This Package

This package ships two integration artifacts for connecting any AI agent to VantaGate:

Artifact File Best for
OpenAPI 3.0 Spec vanta-gate-openapi.json No-code tools, OpenAI GPTs, Alice, n8n, Zapier
MCP Server src/index.ts / dist/index.js Claude Desktop, Cursor, Cline, any MCP-compatible agent

Prerequisites

  • Node.js >= 20
  • A VantaGate account: https://vanta-gate.com
  • A VantaGate API key (Dashboard → Projects → New Project)

Quick Start

Option A: Use the MCP Server with Claude Desktop

The fastest path. No manual setup required.

Step 1: Get your API key from https://vanta-gate.com/dashboard/projects

Step 2: Add VantaGate to your Claude Desktop config.

Open your claude_desktop_config.json:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "vantagate": {
      "command": "npx",
      "args": ["-y", "@vantagate/mcp-server"],
      "env": {
        "VANTA_API_KEY": "YOUR_API_KEY_HERE"
      }
    }
  }
}

Step 3: Restart Claude Desktop. The tools create_vantagate_checkpoint and check_vantagate_status will appear in Claude's tool list.

Step 4: Ask Claude to do something that requires approval:

"Delete all records from the staging database where created_at < 2024-01-01"

Claude will automatically pause, create a checkpoint, and tell you to visit the dashboard_url to approve or reject.


Option B: Use the OpenAPI Spec (No-Code / Low-Code)

Import vanta-gate-openapi.json directly into your tool of choice.

OpenAI GPTs / Custom Actions

  1. Open your GPT editor at https://platform.openai.com/gpts
  2. Click "Add actions" → "Import from URL" or paste the JSON
  3. Set Authentication → API Key → Header: X-API-KEY
  4. Enter your VantaGate API key

The GPT will now pause before high-risk tool calls and ask for human approval.

Alice (AI Agent Platform)

  1. Go to Integrations → Import OpenAPI
  2. Upload or paste vanta-gate-openapi.json
  3. Set header X-API-KEY to your VantaGate API key in the credential store
  4. Map create_vantagate_checkpoint to your agent's "before high-risk action" trigger

n8n / Zapier

  1. Add an HTTP Request node
  2. Import the OpenAPI spec to auto-populate endpoints
  3. Set X-API-KEY header in the credential configuration

LangChain / LlamaIndex

from langchain.tools import OpenAPITool

vanta_tool = OpenAPITool.from_openapi_spec(
    spec_path="./vanta-gate-openapi.json",
    headers={"X-API-KEY": os.environ["VANTA_API_KEY"]}
)
agent = initialize_agent([vanta_tool], llm, agent=AgentType.OPENAI_FUNCTIONS)

Available MCP Tools

create_vantagate_checkpoint

Pauses the agent workflow and routes a human approval request.

When Claude uses it: Before any high-risk action - financial operations, data deletion, production deployments, bulk communications.

Parameter Type Required Description
title string ✅ Short title shown to approver. Max 200 chars.
payload object ✅ Full JSON context for the decision. Encrypted + purged after decision.
description string ❌ Additional context below the title. Max 1000 chars.
options string[] ❌ Decision options. First = approve action. Default: ["Approve", "Reject"]. Min 2, max 5.
timeout string ❌ Auto-expire duration: 30m, 4h, 2d. Default: 24h.
notify_email string ❌ Email address for magic-link notification.
slack_webhook_url string ❌ Slack webhook URL (Pro/Scale plans). From Dashboard → Add to Slack.
callback_url string ❌ Your HTTPS endpoint for signed decision webhook.

Returns: checkpoint_id, dashboard_url, and step-by-step instructions for the agent.

check_vantagate_status

Polls the decision status of a pending checkpoint.

Parameter Type Required Description
checkpoint_id string ✅ The ID from create_vantagate_checkpoint.

Returns: status (PENDING / APPROVED / REJECTED / RESOLVED / EXPIRED), selected_option, reject_reason, and the full audit trail.


API Reference Summary

Base URL: https://api.vanta-gate.com/v1
Authentication: X-API-KEY header

Method Endpoint Description
POST /checkpoint Create a checkpoint - pauses agent
GET /checkpoint/{id} Poll for human decision
GET /checkpoint/secure/{token} Decision UI (internal - magic link)
POST /checkpoint/secure/{token}/decide Submit decision (internal - decision UI)

Checkpoint Status Lifecycle

PENDING → APPROVED   (human chose first/positive option)
        → REJECTED   (human rejected with optional reason)
        → RESOLVED   (human chose non-primary option)
        → EXPIRED    (timeout elapsed, no decision)

Error Codes

HTTP Code Description
400 Invalid_Webhook_URL slack_webhook_url must start with https://hooks.slack.com/
400 Invalid_Callback_URL callback_url is malformed or points to private IP
400 Invalid_Decision Decision value not in checkpoint's options array
400 Validation_Error Request body field validation failure
401 Unauthorized Missing, invalid, or rotated API key
402 Upgrade_Required Feature requires Pro or Scale plan
403 Forbidden API key doesn't match the checkpoint's project
404 Not_Found Checkpoint ID or token does not exist
409 Already_Decided Decision already recorded for this checkpoint
410 Checkpoint_Expired Timeout window has passed
429 Rate_Limited Too many requests - back off and retry

All errors follow the envelope:

{
  "statusCode": 400,
  "error": "Machine_Readable_Code",
  "message": "Human-readable description."
}

Subscription Tiers

Feature Free Pro ($49/mo) Scale ($199/mo)
Checkpoints/month 50 2,500 25,000
Email notifications ✅ ✅ ✅
Slack notifications ❌ ✅ ✅
Webhook callbacks ✅ ✅ ✅
Max timeout 24h 7 days 30 days
Log retention 7 days 90 days 365 days

Timeout values are silently clamped to your plan's maximum. A Free plan request with timeout: "7d" will be capped to 24h.


Webhooks

When a human decides, VantaGate sends a signed HTTPS POST to your callback_url (up to 5 retry attempts with exponential back-off).

Request headers:

Content-Type: application/json
X-Vanta-Signature: sha256=<HMAC-SHA256 of body>
User-Agent: VantaGate-Webhook/1.0

Signature verification (Node.js):

const crypto = require('crypto')

function verifyVantaSignature(rawBody, signature, projectSecret) {
  const expected = 'sha256=' + crypto
    .createHmac('sha256', projectSecret)
    .update(rawBody)
    .digest('hex')
  return crypto.timingSafeEqual(
    Buffer.from(expected),
    Buffer.from(signature)
  )
}

app.post('/webhook/vanta', express.raw({ type: 'application/json' }), (req, res) => {
  const sig = req.headers['x-vanta-signature']
  if (!verifyVantaSignature(req.body, sig, process.env.VANTA_PROJECT_SECRET)) {
    return res.status(401).json({ error: 'Invalid signature' })
  }
  const event = JSON.parse(req.body)
  if (event.status === 'APPROVED') {
    // ✅ Resume agent workflow
  } else if (event.status === 'REJECTED') {
    // ❌ Halt - check event.reject_reason
  }
  res.json({ received: true })
})

Your Webhook Signing Secret (VANTA_PROJECT_SECRET) is distinct from your API key. Find it in Dashboard → Project Settings.


Building from Source

# Install dependencies
npm install

# Build TypeScript
npm run build

# Run the server directly
VANTA_API_KEY=YOUR_API_KEY_HERE npm start

Advanced: MCP Config for Other Clients

Cursor / Cline / Windsurf

{
  "mcpServers": {
    "vantagate": {
      "command": "npx",
      "args": ["-y", "@vantagate/mcp-server"],
      "env": {
        "VANTA_API_KEY": "YOUR_API_KEY_HERE"
      }
    }
  }
}

With local build (development)

{
  "mcpServers": {
    "vantagate": {
      "command": "node",
      "args": ["/absolute/path/to/dist/index.js"],
      "env": {
        "VANTA_API_KEY": "YOUR_API_KEY_HERE"
      }
    }
  }
}

Security

  • Your X-API-KEY is hashed server-side - VantaGate never stores plaintext keys.
  • Your slack_webhook_url is sent per-request and purged atomically after the decision. Zero retention.
  • All payload data is encrypted at rest with AES-256 and destroyed after the human decision. It cannot be reconstructed.
  • callback_url is validated against private IP ranges (SSRF prevention).
  • All webhook deliveries are signed with HMAC-SHA256. Always verify signatures.

License

MIT - see LICENSE


Links

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选