App::MCP::CLI

5 min read
Published 16 Oct 2026
← Back to Blogs
App::MCP::CLI

DISCLAIMER: Image is generated using ChatGPT.


My formal introduction to MCP was by this Dave Cross’s post: Teaching AI About the British Monarchy with MCP

In the article Dave mentioned this line:

MCP is Mostly JSON-RPC

That gave me an idea to create a dedicated blog post: JSON-RPC with Perl.

Fast forward (5 months later), I am back on the subject with another blog post.

This post is mostly about my CPAN distribution, App::MCP::CLI.

In simple term, Model Context Protocol (MCP), is a universal USB port for AI.

It creates a standard plugin that any AI can talk to.

For example, you can create MCP connector for Google Drive and any AI tool that supports MCP can use it to read or search your files.

The inspiration of creating MCP CLI came from this tool: https://pypi.org/project/mcp

I created initial draft few months ago and never got time to revisit it.

Recently I got a reminder about Dave Cross upcoming online talk, MCP - Model Context Protocol in Perl: From Web App to AI Tools

This made me go back and finish it off.

I have an experimental mcpcli ready for you to play.

You just need to install the CPAN module, App::MCP::CLI.

The distribution comes with simple example to let you play with it.

Once you install the distribution, you have everything you need.

Here is a basic stdio server. You would need to install MCP to be able to use it.

This server creates a tool add_numbers which adds two numbers.

There can never be any simpler example than this.

File: stdio_server.pl

#!/usr/bin/env perl

use strict;
use warnings;

use utf8;
use MCP::Server;
use experimental 'signatures';

my $server = MCP::Server->new;

$server->tool(
    name        => 'add_numbers',
    description => 'Adds two numeric values together and returns the sum.',

    input_schema => {
        type       => 'object',
        properties => {
            a => { type => 'number', description => 'The first number'  },
            b => { type => 'number', description => 'The second number' },
        },
        required => ['a', 'b'],
    },

    code => sub ($tool, $args) {
        my $num1 = $args->{a};
        my $num2 = $args->{b};
        my $sum  = $num1 + $num2;

        return "The result of adding $num1 and $num2 is $sum.";
    }
);

$server->to_stdio;

This is how we can all the server:

$ mcpcli call add_numbers --server ./stdio_server.pl a=10 b=5
[-] Calling tool 'add_numbers'...
The result of adding 10 and 5 is 15.

We can make this even more simple to use.

Let’s create server config now:

File: server_config.json

{
  "mcpServers": {
    "perl-math-server": {
      "command": "perl",
      "args": ["stdio_server.pl"]
    }
  },
  "confirmTools": false
}

Now using this config, we can list the servers.

$ mcpcli servers
[-] Reading configuration from: server_config.json

Configured MCP Servers:
+------------------+-----------+----------------------+
| Server Name      | Transport | Command / URL        |
+------------------+-----------+----------------------+
| perl-math-server | stdio     | perl stdio_server.pl |
+------------------+-----------+----------------------+

You can list the tools defined in the server:

$ mcpcli list --server perl-math-server
[-] Reading configuration from: server_config.json
[-] Resolved server settings for 'perl-math-server' via config file.

Available MCP Tools:
+-------------+-------------------------------------------------------+
| Tool Name   | Description                                           |
+-------------+-------------------------------------------------------+
| add_numbers | Adds two numeric values together and returns the sum. |
+-------------+-------------------------------------------------------+

Now the above mcp call becomes this:

$ mcpcli call add_numbers --server perl-math-server a=10 b=5
[-] Reading configuration from: server_config.json
[-] Resolved server settings for 'perl-math-server' via config file.
[-] Calling tool 'add_numbers'...
The result of adding 10 and 5 is 15.

This is much better, in my opinion.

How about make it interactive?

We would need to install Ollama first.

I am using Docker as I am in love with it.

$ docker run -d -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama

Then pull the model qwen2.5:1.5b, the default model used by mcpcli.

This can be overriden by the switch --model flag.

$ docker exec -it ollama ollama pull qwen2.5:1.5b

Verify if the model in installed:

$ docker exec -it ollama ollama list

Once you have the setup ready.

We can now start the interactive session as below:

$ mcpcli chat --server perl-math-server
[-] Reading configuration from: server_config.json
[-] Resolved server settings for 'perl-math-server' via config file.
[-] Launching Local Pipe MCP Server: perl stdio_server.pl
[+] Connected to MCP Server. Found 1 tool(s).
[-] Connected to LLM provider 'ollama' using model 'qwen2.5:1.5b'.
[-] Starting agent loop. Type 'exit' or 'quit' to end session.

You>

Now you can type the message as below:

You> What is 10 plus 5?
Thinking...
 -> [Agent decided to call tool 'add_numbers' with args: {"b":5,"a":10}]
Response: The sum of 10 and 5 is 15.

You>

Isn’t fun?

It is all happening using your favourite programming language Perl.

Next example is SSE after we have done with stdio.

Source: https://github.com/manwar/App-MCP-CLI/blob/master/examples/sse_server.pl

This is little tricky but not that complicated.

The server by default runs on 127.0.0.1 and listen to port 8080.

You can override that using env vars SSE_HOST and SSE_PORT.

You start the server as below:

$ perl sse_server.pl
[-] Mock MCP SSE Server listening at: http://127.0.0.1:8080/sse
[-] Press Ctrl+C to stop the server

Now we actual test the sse server, let’s update the server_config.json first.

{
  "mcpServers": {
    "perl-math-server": {
      "command": "perl",
      "args": ["stdio_server.pl"]
    },
    "weather-sse-service": {
      "transport": "sse",
      "url": "http://localhost:8080/sse"
    }
  },
  "confirmTools": false
}

Let’s list all the servers now:

$ mcpcli servers
[-] Reading configuration from: server_config.json

Configured MCP Servers:
+---------------------+-----------+---------------------------+
| Server Name         | Transport | Command / URL             |
+---------------------+-----------+---------------------------+
| perl-math-server    | stdio     | perl stdio_server.pl      |
| weather-sse-service | SSE       | http://localhost:8080/sse |
+---------------------+-----------+---------------------------+

List MCP tools supported by the SSE server:

$ mcpcli list --server weather-sse-service
[-] Reading configuration from: server_config.json
[-] Resolved server settings for 'weather-sse-service' via config file.

Available MCP Tools:
+------------------+---------------------------------------------------------+
| Tool Name        | Description                                             |
+------------------+---------------------------------------------------------+
| get_mock_weather | Returns static mock weather information for a location. |
+------------------+---------------------------------------------------------+

Finally start the interactive session with SSE server as below:

$ mcpcli chat --server weather-sse-service
[-] Reading configuration from: server_config.json
[-] Resolved server settings for 'weather-sse-service' via config file.
[-] Connecting to MCP Server via HTTP/SSE: http://localhost:8080/sse
[-] SSE Channel Established. Target Post Route: http://127.0.0.1:8080/message
[+] Connected to MCP Server. Found 1 tool(s).
[-] Connected to LLM provider 'ollama' using model 'qwen2.5:1.5b'.
[-] Starting agent loop. Type 'exit' or 'quit' to end session.

You>

Now you type the message:

You> what is the weather in London?
Thinking...
 -> [Agent decided to call tool 'get_mock_weather' with args: {"location":"London"}]
Response: The current temperature in London is 21 degrees Celsius and there are clear skies.

You>

Happy Hacking !!!

Tags MCP