Build Multi-Agent MCP Systems in Browser with WASM

Agent MCP Studio is a browser-only IDE for building and orchestrating MCP (Model Context Protocol) agent systems. The entire stack — tool authoring, multi-agent orchestration, RAG, code execution — runs from a single static HTML file via WebAssembly. No backend, no Docker, no server. Close the tab and everything is gone.

Key Features

WASM-based sandbox: Tools run in Pyodide (Python) or DuckDB-WASM (SQL), with no code leaving the browser. LLM-generated code is AST-validated before registration, then JIT-compiled on first call.
10 orchestration strategies: Supervisor, Mixture of Experts (parallel + synthesizer), Sequential Pipeline, Plan & Execute, Swarm, Debate, Reflection, Hierarchical, Round-Robin, Map-Reduce. You drag tool chips onto persona nodes on a service graph, pick a strategy, and the topology adapts.
On-device RAG: Uses Xenova/all-MiniLM-L6-v2 via Transformers.js for local embeddings. No network calls.
Built-in LLM: Supports OpenAI Chat Completions or a local Qwen 1.5 0.5B running in-browser via Transformers.js for fully offline mode.
Bridge.js: A Node bridge speaks stdio to Claude Desktop and WebSocket to your tab — so your browser becomes an MCP server.
Export to production: Generates a real Python MCP server: server.py, agentic.py (faithful port of the browser orchestration), tools/*.py, Dockerfile, requirements.txt, .env.example. Also exports as a single .agentpack.json (Project Pack), which auto-detects required external services from os.environ.get(...) calls.

Practical Commands

# Build the exported Docker image
cd agent-mcp-server-YYYY-MM-DD
docker build -t agent-mcp-export .

Run (stdio – for local use with Claude Desktop)
docker run --rm -i 
  -e MCP_ALLOWED_HOSTS='api.github.com,*.githubusercontent.com' 
  agent-mcp-export
Wire Claude Desktop to the container
{
  "mcpServers": {
    "agent-mcp-studio-export": {
      "command": "docker",
      "args": ["run","--rm","-i","-e","MCP_ALLOWED_HOSTS=api.github.com","agent-mcp-export"]
    }
  }
}

The export also supports deploying to Fly.io, Railway, Render, Cloud Run, ECS, etc. The image defaults to stdio; for HTTP, see the README.

Who it's for: Developers building multi-agent systems who want a zero-infrastructure prototype-to-production pipeline, especially those experimenting with MCP and local-first AI.

📖 Read the full source: HN AI Agents