@alfe.ai/mcp-bundler

Provider-agnostic MCP server bundler — connects to N MCP servers (stdio/SSE/streamable-http), aggregates their tools with namespacing, exposes a unified callable surface

Part of Alfe — the operating system for AI agents: build, deploy, and run agents with persistent memory, identity, integrations, and channels. See the documentation to get started.

Install

npm install @alfe.ai/mcp-bundler

Runtime model

Store is the owner-only on-disk source of truth, Manager serializes configuration mutations and reconciles them into a live McpBundler, and the bundler supervises child connections while exposing one collision-safe tool catalogue. Tool listing and routing use the same resolved catalogue, so every advertised name is callable.

The store and child MCP protocol are executable-data boundaries: configuration, catalogues, schemas, arguments, results, stderr, and errors are validated and bounded. Existing malformed stores fail closed instead of being replaced with an empty registry. Store files are read without following symlinks and are tightened to owner-only permissions.

import { Manager, McpBundler } from '@alfe.ai/mcp-bundler';

const bundler = new McpBundler();
const manager = new Manager();
await manager.loadIntoBundler(bundler);
await manager.addServer(
  { command: 'npx', args: ['-y', 'example-mcp'] },
  { id: 'example', owner: 'manual' },
);

Links

• 🌐 Website: https://alfe.ai
• 📚 Docs: https://docs.alfe.ai