Pyjo-MCP

Model Context Protocol server and client for Pyjoyment, of the revision 2026-07-28 of the specification.

Installation

From PyPI, with Pyjoyment, which it depends on:

pip install Pyjo-MCP

or, in a project managed with uv:

uv add Pyjo-MCP

The development version is installed from the Git repository:

pip install git+https://github.com/pyjoyment/Pyjo-MCP

Related MCP server: MCP Python Server

Usage

A server of MCP is a route of an application of Pyjoyment, whose tools are functions, with the schemas of their type hints and the descriptions of their docstrings:

from Pyjo.MCP.Server import Server
from Pyjo.MCP.Tool import Error
from Pyjoyment.Lite import *

server = Server(name='calculator', version='1.0.0')


@server.tool
def divide(a: float, b: float) -> float:
    """Divide the first number by the second one."""
    if b == 0:
        raise Error.new('Division by zero')
    return a / b


# Host and Origin of localhost only, against DNS rebinding
plugin('AllowedHosts')

any('/mcp', server.to_action())

app.start()

uv run python calculator.py daemon -l http://127.0.0.1:3000

Its endpoint is http://127.0.0.1:3000/mcp, of the transport Streamable HTTP. The whole calculator is examples/calculator.py.

Prompts, the templates of messages that a user picks, and resources, the data that a client reads by its URI, are functions too:

@server.prompt
def explain(expression: str) -> str:
    """Explain an arithmetic expression step by step."""
    return f'Explain step by step how to compute {expression}.'


@server.resource('calculator://precision', mime_type='application/json')
def precision() -> dict:
    """The precision of the numbers of the calculator."""
    return {'digits': 15}

The same application serves the transport stdio too, with a hook:

hook('before_server_start', server.to_stdio())

A client of MCP, as Claude Code, then starts it with the command stdio:

claude mcp add calculator -- uv run python examples/calculator.py stdio

examples/weather.py is a gateway to the weather of Open-Meteo, which needs no key, with the tools find_places, current_weather, forecast and hourly_forecast, which request the API with the non-blocking user agent of the application. A client of MCP, as Claude Code, connects to it over HTTP:

uv run python examples/weather.py daemon -l http://127.0.0.1:3000
claude mcp add --transport http weather http://127.0.0.1:3000/mcp

examples/mcp_cli.py is a client of the command line for the tools of any server of MCP: a tool is a command, in kebab-case, and its arguments are options, with the descriptions and the constraints of its inputSchema in the help of the command. The result is text, with tables of the lists of objects, or JSON:

uv run python examples/mcp_cli.py --help
uv run python examples/mcp_cli.py forecast --place Kraków --language pl --days 2
uv run python examples/mcp_cli.py hourly-forecast -p Warsaw -H 6 -f json

It is made with Pyjo.MCP.Client, a client of MCP over Streamable HTTP, which sends the _meta and the headers that every request carries:

from Pyjo.MCP.Client import Client

client = Client('http://127.0.0.1:3000/mcp')
print(client.call_tool('divide', {'a': 7, 'b': 2})['structuredContent'])

Pyjo.MCP.Test tests a server with the requests of the client, without starting it:

import Pyjo.MCP.Test

from Pyjo.File import curfile
from Pyjo.Test import *

t = Pyjo.MCP.Test.new(curfile().sibling('calculator.py'))
t.call_ok('divide', a=7, b=2).status_is(200).structured_content_is(3.5)
t.call_ok('divide', a=1, b=0).is_error_is(True).text_content_is('Division by zero')

done_testing()

Documentation

The tutorial, which makes servers of MCP step by step, and the API of the package are on Read the Docs.

Status

A server of the tools, the prompts and the resources of MCP over Streamable HTTP and stdio, with server/discover, tools/list, tools/call, prompts/list, prompts/get, resources/list, resources/templates/list, resources/read, subscriptions/listen and completion/complete, a client of them over Streamable HTTP, and assertions of tests with its requests. A tool sends its progress and its log messages before its result, and a subscription gets the changes of the lists and of the resources. A resource template is a URI Template of RFC 6570, whose URIs a client reads. A tool, a prompt and a resource ask the user, the language model or the roots of the client for input, in the rounds of a multi round-trip request, and the arguments of the prompts and the variables of the templates complete their values.

License

The Artistic License 2.0, as Pyjoyment.