The AI agent multiplexer. A self-hosted platform for creating, orchestrating, and managing teams of AI agents through a web interface.

muxAI lets you build multi-agent teams that collaborate on tasks. Define agents with different roles, group them into hierarchical teams, and watch them work together in real-time. Agents communicate through MCP (Model Context Protocol) servers, capture structured results, and can even make autonomous payments via x402.

Key capabilities:

• Multi-agent teams with lead/reporter hierarchy and parallel orchestration
• Built-in MCP servers for news analysis, chart analysis, crypto data, OHLCV indicators, and more
• Real-time streaming of agent activity via SSE
• Structured result capture with customizable result cards
• Contractor models for cross-provider agent invocation (OpenRouter, GPT-4o, etc.)
• x402 payments with auto-generated Solana + Base wallets per agent
• Heartbeat scheduling for recurring agent runs via cron
• Chat interface for conversational interaction with agents
• Team templates for one-click deployment of pre-configured agent teams

muxai-io/
├── apps/
│   ├── api/                Express backend (port 3001)
│   │   ├── prisma/         Schema + migrations
│   │   └── src/
│   │       ├── routes/     API endpoints
│   │       ├── services/   Business logic, adapters, spawn
│   │       └── lib/        Prisma client, utilities
│   └── web/                Next.js frontend (port 3000)
│       └── src/
│           ├── app/        Pages and routes
│           ├── components/ UI components
│           └── lib/        Templates, utilities, types
├── packages/
│   ├── mcp-orchestrator/   run_team + ask_reporter tools
│   ├── mcp-news-analyst/   RSS + CMC news aggregation
│   ├── mcp-chart-analyst/  Chart image analysis
│   ├── mcp-crypto-ohlcv/   OHLCV candles + technical indicators
│   ├── mcp-crypto-data/    Funding rates, OI, L/S ratio, taker volume
│   ├── mcp-wallet/         x402 wallet tools (address, fetch-and-pay)
│   ├── mcp-contractor/     Cross-provider model invocation
│   └── mcp-docs/           Mintlify documentation proxy
├── config/
│   └── mcp-registry.json   Built-in MCP server registry
├── install/                Install + update scripts
├── docker-compose.yml      Development database
└── docker-compose.test.yml Test database

For full installation and update instructions, see the muxAI Quickstart Guide.

The guide covers prerequisites, install scripts, manual setup, configuration, and running the platform.

1. Create an agent via the UI with a role, model, system prompt, and MCP configuration
2. Run on demand or via heartbeat schedule — the API spawns a Claude CLI process
3. Stream logs in real-time via SSE to the frontend
4. Capture structured results (trade decisions, analysis, etc.) from agent output
5. Orchestrate teams where a lead agent delegates to reporters via MCP tools

Each agent runs with strict MCP configuration. Built-in servers are resolved from the registry, reporter agents are excluded from orchestrator tools (preventing recursive spawning), and custom MCP servers from the database are merged at runtime.

Teams use a lead/reporter pattern:

• The lead agent has access to run_team (parallel broadcast) and ask_reporter (targeted follow-up) MCP tools
• Reporter agents run independently with their own MCP tools scoped to their role
• The lead synthesizes reporter outputs into a final result

muxAI has two test suites: unit tests and integration tests.

# Unit tests (no database required)
pnpm test

# Integration tests (requires test database)
docker compose -f docker-compose.test.yml up -d
cd apps/api && pnpm test:integration

See TESTING.md for full details on the test infrastructure, writing new tests, and CI setup.

MIT