AgentGate

Human-in-the-loop approval system for AI agents.
Agents request. Policies decide. Humans approve.
Keep humans in control of what AI agents can do.

Your AI agent wants to send an email, delete a file, or deploy to production. Should it? AgentGate lets you define policies that auto-approve safe actions, auto-deny dangerous ones, and route everything else to a human — via dashboard, Slack, Discord, or email.

✨ Highlights

• 🛡️ Policy engine — auto-approve, auto-deny, or route to humans based on rules
• 👥 Multi-channel approvals — Slack, Discord, email, or web dashboard
• 🔌 TypeScript SDK + MCP — works with any agent framework or Claude Desktop
• 🪝 Webhooks with retry — real-time notifications with exponential backoff
• 📝 Full audit trail — every request, decision, and action logged
• 🐳 Docker-ready — one docker-compose up for the full stack
• 🔐 Production-hardened — SSRF protection, ReDoS defense, structured logging, graceful shutdown
• ⚡ Fast & lightweight — Hono server, SQLite or PostgreSQL

Table of Contents

• Quick Start
• Architecture
• Packages
• SDK Usage
• CLI
• MCP Integration
• Authentication
• API Endpoints
• Rate Limiting
• Webhooks
• Configuration
• Docker Deployment
• Development
• Contributing
• License

Quick Start

1. Install dependencies

pnpm install

2. Run database migrations

pnpm --filter @agentgate/server db:migrate

3. Bootstrap (create admin API key)

pnpm --filter @agentgate/server bootstrap

Save the API key - it's shown once only! Set it in your environment:

export AGENTGATE_API_KEY="agk_..."

4. Start the development environment

# Start server (port 3000) and dashboard (port 5173)
pnpm dev

5. Run the demo

In a new terminal (with API key set):

export AGENTGATE_API_KEY="agk_..."
pnpm demo

6. Open the dashboard

Visit http://localhost:5173 to view and manage approval requests.

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                         AI Agents                                │
│  (use @agentgate/sdk or MCP to request approvals)               │
└───────────────────────────┬─────────────────────────────────────┘
                            │ HTTP API (authenticated)
                            ▼
┌─────────────────────────────────────────────────────────────────┐
│                    AgentGate Server                              │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐          │
│  │ Policy Engine│  │ Request Store│  │ Audit Logger │          │
│  ├──────────────┤  ├──────────────┤  ├──────────────┤          │
│  │  API Keys    │  │  Webhooks    │  │  MCP Server  │          │
│  └──────────────┘  └──────────────┘  └──────────────┘          │
└───────────────────────────┬─────────────────────────────────────┘
                            │
              ┌─────────────┼─────────────┐
              ▼             ▼             ▼
┌────────────────┐ ┌────────────────┐ ┌────────────────┐
│  Web Dashboard │ │   Slack Bot    │ │  Discord Bot   │
│(React+Tailwind)│ │(approve in DM) │ │(approve in ch) │
└────────────────┘ └────────────────┘ └────────────────┘
              │             │             │
              └─────────────┼─────────────┘
                            ▼
                      ┌──────────┐
                      │  Humans  │
                      └──────────┘

Packages

SDK Usage

import { AgentGateClient } from '@agentgate/sdk';

// Create client with API key
const client = new AgentGateClient({
  baseUrl: 'http://localhost:3000',
  apiKey: process.env.AGENTGATE_API_KEY,
});

// Request approval
const request = await client.request({
  action: 'send_email',
  params: {
    to: 'customer@example.com',
    subject: 'Order shipped!',
  },
  urgency: 'normal',
});

// Wait for human decision
const decided = await client.waitForDecision(request.id, {
  timeout: 60000, // 1 minute
});

if (decided.status === 'approved') {
  // Execute the action
  await sendEmail(decided.params);
} else {
  console.log('Action denied:', decided.decisionReason);
}

CLI

AgentGate includes a command-line interface for managing approval requests.

Installation

# From the monorepo
pnpm --filter @agentgate/cli build

# Or install globally (when published)
npm install -g @agentgate/cli

Configuration

Configure the CLI with your server URL and API key:

# Set server URL
agentgate config set serverUrl http://localhost:3000

# Set API key
agentgate config set apiKey agk_your_api_key

# View current config
agentgate config show

Configuration is stored in ~/.agentgate/config.json. You can also use environment variables:

export AGENTGATE_URL=http://localhost:3000
export AGENTGATE_API_KEY=agk_...

Commands

Examples

# Create a request
agentgate request send_email \
  --params '{"to": "user@example.com", "subject": "Hello"}' \
  --urgency high

# List pending requests
agentgate list --status pending

# Approve a request
agentgate approve req_abc123 --reason "Looks good"

# Deny a request
agentgate deny req_abc123 --reason "Not authorized"

# Output as JSON
agentgate list --json

MCP Integration

AgentGate includes a Model Context Protocol (MCP) server for integration with Claude Desktop and other MCP-compatible clients.

Claude Desktop Configuration

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "agentgate": {
      "command": "npx",
      "args": ["@agentgate/mcp"],
      "env": {
        "AGENTGATE_URL": "http://localhost:3000",
        "AGENTGATE_API_KEY": "agk_..."
      }
    }
  }
}

Available MCP Tools

Authentication

AgentGate uses API keys for authentication. All API requests (except /health) require a valid API key.

API Key Scopes

Using API Keys

HTTP Header:

curl -H "Authorization: Bearer agk_..." http://localhost:3000/api/requests

SDK:

const client = new AgentGateClient({
  baseUrl: 'http://localhost:3000',
  apiKey: process.env.AGENTGATE_API_KEY,
});

Creating Additional API Keys

// Via API (requires admin scope)
POST /api/keys
{
  "name": "My Agent",
  "scopes": ["request:create", "request:read"]
}

API Endpoints

Rate Limiting

AgentGate supports per-API-key rate limiting to prevent abuse and ensure fair usage.

How It Works

• Rate limits use a sliding window algorithm (requests per minute)
• Limits are configured per API key
• When exceeded, requests return 429 Too Many Requests
• Rate limit headers are included in all authenticated responses

Rate Limit Headers

Configuring Rate Limits

Set rate limits when creating or updating API keys:

// Via API (requires admin scope)
POST /api/keys
{
  "name": "My Agent",
  "scopes": ["request:create", "request:read"],
  "rateLimit": 60  // 60 requests per minute
}

// null = unlimited
{
  "name": "Internal Service",
  "scopes": ["admin"],
  "rateLimit": null
}

Dashboard

Rate limits can also be managed from the web dashboard under Settings → API Keys.

Webhooks

AgentGate can notify external systems when request events occur.

Setting Up Webhooks

// Create a webhook via API
POST /api/webhooks
{
  "url": "https://your-server.com/webhook",
  "events": ["request.created", "request.decided"],
  "secret": "optional-signing-secret"
}

Webhook Events

Webhook Payload

{
  "event": "request.decided",
  "timestamp": "2024-01-15T10:30:00Z",
  "data": {
    "id": "abc123",
    "action": "send_email",
    "status": "approved",
    "decidedBy": "admin@example.com"
  }
}

Webhook Signatures

If you provide a secret, requests are signed with HMAC-SHA256:

X-AgentGate-Signature: sha256=...

Verify by computing HMAC-SHA256(secret, body) and comparing.

Webhook Retry

Failed webhook deliveries are retried automatically with exponential backoff. The server scans for pending deliveries and retries them with increasing delays (2^attempts * 1000ms) until successful or the maximum retry count is reached.

Configuration

Environment Variables

Policy Configuration

Policies are stored in the database and can be managed via API:

// Example: Auto-approve low-risk emails
{
  name: "auto-approve-emails",
  priority: 10,
  enabled: true,
  rules: [
    {
      match: { action: "send_email" },
      decision: "auto_approve"
    }
  ]
}

Docker Deployment

AgentGate provides Docker images for easy self-hosted deployments.

Quick Start

1. Copy the example environment file:

cp .env.example .env

2. Generate secure credentials:

# Generate admin API key (required)
echo "ADMIN_API_KEY=$(openssl rand -hex 32)" >> .env

# Generate JWT secret (recommended for production)
echo "JWT_SECRET=$(openssl rand -hex 32)" >> .env

3. Start all services:

docker-compose up -d

4. Access the services:

• Dashboard: http://localhost:8080
• API Server: http://localhost:3000
• Health Check: http://localhost:3000/health

Services

With Slack or Discord Bots

To include the bot services, use the bots profile:

# Set required bot credentials in .env first
docker-compose --profile bots up -d

Configuration

All configuration is done via environment variables. See .env.example for all options.

Required variables:

• ADMIN_API_KEY — Admin API key (min 16 characters)

Recommended for production:

• JWT_SECRET — JWT signing secret (min 32 characters)
• CORS_ORIGINS — Restrict to your domain(s)
• POSTGRES_PASSWORD — Use a strong password

Building Images

Build all images locally:

docker-compose build

Build a specific service:

docker-compose build server
docker-compose build dashboard

Database Migrations

Migrations run automatically when the server starts. For manual control:

# Run migrations inside the container
docker-compose exec server node -e "
  import('./dist/db/migrate.js').then(m => m.runMigrations())
"

Viewing Logs

# All services
docker-compose logs -f

# Specific service
docker-compose logs -f server

# Last 100 lines
docker-compose logs --tail=100 server

Stopping Services

# Stop all
docker-compose down

# Stop and remove volumes (WARNING: deletes data)
docker-compose down -v

Production Considerations

1. Use a reverse proxy (nginx, Caddy, Traefik) for TLS termination
2. Set strong passwords for PostgreSQL
3. Restrict CORS origins to your domain
4. Use Docker secrets for sensitive values in production
5. Set up backups for PostgreSQL data volume
6. Monitor health endpoints for uptime checks

Development

# Install dependencies
pnpm install

# Run migrations
pnpm --filter @agentgate/server db:migrate

# Bootstrap (create admin key)
pnpm --filter @agentgate/server bootstrap

# Start development (server + dashboard)
pnpm dev

Testing

AgentGate uses Vitest for testing across all packages.

# Run all tests
pnpm test

# Run tests with coverage report
pnpm test:coverage

# Run tests in watch mode (single package)
pnpm --filter @agentgate/server test:watch

# Run a specific test file
pnpm --filter @agentgate/server test -- src/__tests__/integration.test.ts

Coverage reports are generated per-package and include line, branch, and function coverage.

Code Quality

# Build all packages
pnpm build

# Type checking
pnpm typecheck

# Lint (ESLint)
pnpm lint

# Fix lint issues
pnpm lint:fix

# Format code (Prettier)
pnpm format

# Check formatting
pnpm format:check

Project Structure

agentgate/
├── packages/
│   ├── core/           # Shared types, schemas, policy engine
│   ├── server/         # Hono API server
│   ├── sdk/            # TypeScript SDK
│   ├── cli/            # Command-line interface
│   ├── mcp/            # MCP server for Claude Desktop
│   ├── slack/          # Slack bot
│   ├── discord/        # Discord bot
│   └── dashboard/      # React dashboard
├── apps/
│   └── demo/           # Demo application
├── docker-compose.yml  # Docker deployment
└── package.json        # Monorepo root

See Also

FormBridge — Mixed-mode forms for AI agents and humans. If AgentGate handles action approval, FormBridge handles data collection. Together they cover the full spectrum of human-AI collaboration: agents gather data with human help (FormBridge), then request permission to act on it (AgentGate).

Contributing

Contributions are welcome! To get started:

1. Fork the repository
2. Clone and install dependencies (pnpm install)
3. Follow the Development section above to set up your local environment
4. Create a feature branch and make your changes
5. Run pnpm build && pnpm test to verify everything works
6. Open a pull request

Please make sure all tests pass and code is formatted (pnpm format:check && pnpm lint) before submitting.

License

MIT