Installation

Milford ships as npm packages under the @milfordai scope and as Docker images. Pick the server to run flows over HTTP, the MCP server to expose flows as tools, or the libraries to embed the engine in your own app.

Requirements

Node.js 22 or later. Every package declares this in engines.
npm or pnpm.
Docker, if you run the images.

Packages

Package	Install it to	Provides
`@milfordai/server`	Run flows over HTTP. Includes Slack, Telegram and webhook channels.	the `milford-server` command
`@milfordai/mcp`	Expose flows as tools to LLM clients, and call other MCP servers.	the `milford-mcp` command
`@milfordai/core`	Embed the engine in a TypeScript app.	flows, nodes and the provider port. No I/O.
`@milfordai/providers`	Add OpenAI-compatible servers, Anthropic, Jev and generic HTTP providers to the library.	provider adapters
`@milfordai/config`	Load a config file yourself.	the config loader and the JSON Schema
`@milfordai/channels`	Use the Slack, Telegram and webhook adapters directly.	channel adapters

The server and MCP packages already depend on the others, so installing one of them is enough to run Milford. All packages are released together. Keep every @milfordai/* dependency on the same version.

Run a server

Run the server without installing anything:

npx @milfordai/server milford.config.yaml

pnpm dlx @milfordai/server milford.config.yaml

Run the MCP server the same way with @milfordai/mcp. To add the server to a project, install it and run the command it provides:

npm install @milfordai/server
npx milford-server milford.config.yaml

pnpm add @milfordai/server
pnpm exec milford-server milford.config.yaml

To put milford-server and milford-mcp on your PATH, install the packages globally:

npm install --global @milfordai/server @milfordai/mcp

pnpm add --global @milfordai/server @milfordai/mcp

The commands read the config file you pass, or the path in MILFORD_CONFIG, or milford.config.yaml in the current directory. See the configuration reference.

Use the library

npm install @milfordai/core @milfordai/providers

pnpm add @milfordai/core @milfordai/providers

The packages are ES modules and ship their own type declarations, so use import and no extra @types package.

import { createEngine, defaultRegistry, flow } from "@milfordai/core";
import { registerProviders } from "@milfordai/providers";

The quickstart has a complete example.

Editor completion for the config file

@milfordai/config ships the JSON Schema of the config file. After you install it, point your editor’s YAML language server at it with the first line of milford.config.yaml:

# yaml-language-server: $schema=./node_modules/@milfordai/config/milford.config.schema.json

Docker

Image	Runs
`ghcr.io/milfordai/milford`	The HTTP server
`ghcr.io/milfordai/milford-mcp`	The MCP server. It needs `mcp.transport: http` in the config.

Tags are the release version, such as 0.0.3, and latest. Mount a directory with milford.config.yaml at /config, and pass the variables your config uses:

docker run -p 8080:8080 \
  -e MILFORD_TOKEN=change-me \
  -v "$PWD:/config:ro" \
  ghcr.io/milfordai/milford:latest

Images are built for linux/amd64 and linux/arm64. See deployment.

Next steps

Quickstart

Run your first flow.

Configuration

Every key of the config file.

Get started

Concepts

Building blocks

Guides

Reference

API reference

Requirements

Packages

Run a server

Use the library

Editor completion for the config file

Docker

Next steps

Quickstart

Configuration

​Requirements

​Packages

​Run a server

​Use the library

​Editor completion for the config file

​Docker

​Next steps

Quickstart

Configuration

Requirements

Packages

Run a server

Use the library

Editor completion for the config file

Docker

Next steps