Following the exploration of FastMCP
, let's turn the attention to another promising gem for building Model Context Protocol (MCP) servers with Ruby on Rails: ActionMCP
. This library is designed with deep Rails integration in mind and offers a robust solution for exposing your application's capabilities to AI agents.
ActionMCP gem
The ActionMCP
gem, developed by seuros (who you might find on the Ruby AI Builders or Model Context Protocol Discord servers), provides a Rails-centric approach to implementing MCP. You can find its repository at https://github.com/seuros/action_mcp, and there's even a helpful Rails template to get you started: https://github.com/seuros/mcp_rails_template.
One of the unique aspects of ActionMCP
is its implementation of the streamable-http
transport method, which aligns with the latest MCP specifications.
Setting Up the Application
To begin, I generated a new Rails application, specifying PostgreSQL as the database:
rails _8.0.2_ new with-action-mcp -d postgresql
It's important to have a working database for your project because ActionMCP
utilizes a database-centered approach for session handling. As outlined in the MCP specification - client has to establish one session per MCP server.
Adding ActionMCP
All the steps needed to install and run ActionMCP
are described in the gem’s repository. I’ve started by adding the actionmcp
gem to the project's Gemfile:
$ bundle add actionmcp
Configuring the gem
With the gem installed, the next step was to instal migrations
$ bin/rails action_mcp:install:migrations
Copied migration 20250517154542_consolidated_migration.action_mcp.rb from action_mcp
Then set up the database and run migrations
$ rails db:setup
$ rails db:migrate
With that finished, I ran the gem’s installation generator:
$ bin/rails generate action_mcp:install
Which created the following files:
create app/mcp/prompts/application_mcp_prompt.rb
create app/mcp/tools/application_mcp_tool.rb
create app/mcp/resource_templates/application_mcp_res_template.rb
create config/mcp.yml
These files provide the basic structure for defining your MCP prompts, tools, and resource templates.
Rackup File and Application Configuration
To be able to run the ActionMCP engine, I created a mcp.ru
file in the root of my Rails application with the following content:
# Load the full Rails environment to access models, DB, Redis, etc.
require_relative "config/environment"
# No need to set a custom endpoint path. The MCP endpoint is always served at root ("/")
# when using ActionMCP::Engine directly.
run ActionMCP::Engine
Next, I added some configuration to config/application.rb
:
config.action_mcp.name = "Friendly MCP (Master Control Program)"
config.action_mcp.version = "1.2.3"
config.action_mcp.logging_enabled = true
config.action_mcp.logging_level = :info
config.action_mcp.vibed_ignore_version = false
For real-time communication, ActionMCP
can leverage different pub/sub adapters. I configured it to use solid_cable
by adding the following to config/mcp.yml
:
development:
adapter: solid_cable
polling_interval: 0.1.seconds
Creating a Weather Forecast Tool
With the basic setup complete, it was time to create a simple tool. I used the provided generator to create a ForecastWeatherTool
:
$ bin/rails generate action_mcp:tool ForecastWeatherTool
This created the file app/mcp/tools/forecast_weather_tool.rb
. I then filled it with dummy logic:
# frozen_string_literal: true
class ForecastWeatherTool < ApplicationMCPTool
tool_name "forecast-weather"
description "Forecasts weather for the given location"
property :location, type: "string", description: "Location for the weather forecast", required: true
def perform
render(text: "Weather in #{location} will be fine")
rescue => e
render(error: [ "Error: #{e.message}" ])
end
end
This tool defines its name, description, and a required location
property. The perform
method contains the core logic, in this case, rendering a simple string in the tool’s response.
Running and testing the server
As per the ActionMCP
documentation, I started the MCP server using a specific command that points to the mcp.ru
file:
bin/rails s -c mcp.ru -p 62770 -P tmp/pids/mcps0.pid
I then launched the MCP Inspector:
npx @modelcontextprotocol/inspector
Connecting the inspector to http://localhost:62770
(the port I specified for the server) and setting streamable HTTP
transport type, I could see and interact with my ForecastWeatherTool
.
Here’s how it appeared in the MCP Inspector:

A More Railsy Example: Creating Blog Posts
To demonstrate an example similar to Rails tutorial’s use case, I created a tool adding blog posts. First, I generated a Post
model with title
and body
attributes (e.g., bin/rails g model Post title:string body:text
).
Then, I generated a new tool for creating posts:
$ bin/rails generate action_mcp:tool CreatePostTool
And implemented it in app/mcp/tools/create_post_tool.rb
:
# frozen_string_literal: true
class CreatePostTool < ApplicationMCPTool
tool_name "create-post"
description "Creates a post with the given title and body"
property :title, type: "string", description: "Title of the post", required: true
property :body, type: "string", description: "Body of the post", required: true
def perform
post = Post.create!(title:, body:)
render(text: "Post #{title} was created")
rescue => e
render(error: [ "Error: #{e.message}" ])
end
end
This tool takes title
and body
as inputs, creates a Post
record, and returns a confirmation message.
After restarting the MCP server, I connected to it again with the MCP Inspector and called the create-post
tool.
Here’s the result in the inspector:

A quick check using the Rails console confirmed that the post was successfully created:
with-action-mcp(dev)> Post.last
Post Load (1.1ms) SELECT "posts".* FROM "posts" ORDER BY "posts"."id" DESC LIMIT 1 /*application='WithActionMcp'*/
=>
#<Post:0x000000011fd0ee70
id: 1,
title: "Ruby on Rails taking over the world",
body: "The post needs some body",
created_at: "2025-05-21 17:55:21.274883000 +0000",
updated_at: "2025-05-21 17:55:21.274883000 +0000">
Summary
ActionMCP
offers a mature, production-ready solution for building MCP servers in a Rails environment. Its sole reliance on the streamable-http
transport type adheres closely to the latest MCP specifications, making it a forward-looking choice.
One current consideration is that not all MCP clients may fully support this advanced transport method yet, as many existing clients are built around stdio
and/or sse
. However, as the MCP ecosystem evolves, broader support for streamable-http
is surely coming.
Overall, ActionMCP
presents a compelling option for Rails developers looking to integrate their applications with AI systems, and I'm looking forward to seeing how it develops and how it can be leveraged in future projects.