An agent doesn't have to guess. Everything is written for it.

An agent writes good code on a framework it can read: its classes, its rules, its traps, and a command that says when the work is done. MMOFramework gives it all four.

Connect your agent

The documentation and the API of every library are served by an MCP server. With Claude Code, one line:

claude mcp add --transport http mmoframework https://www.mmoframework.fr/mcp.json

Any other MCP client takes the same server in its configuration:

{
    "mcpServers": {
        "mmoframework": {
            "type": "http",
            "url": "https://www.mmoframework.fr/mcp.json"
        }
    }
}

What it can ask

search_documentation: Finds a class, a function, an element or a guide across every library, from the words of the task.
documentation_index: Lists what a library documents, to read the right page rather than guess.
describe_class: Gives a class's methods, their parameters and what they return.
describe_function: Gives a function's signature and its documentation.
list_packages: Lists every library, its version, its dependencies and its documents.

What every project carries for it

A project created by the framework comes with the files an agent needs, next to its code:

AGENTS.md: The project's layout, its rules and its commands: what an agent reads first.
CLAUDE.md: Points Claude Code at AGENTS.md, so every agent reads the same rules.
.mcp.json: Connects the agent to the project's own MCP server in development: its routes, entities and classes, live.
.claude/skills: How to write PHP and .jhs on the framework, and its traps confirmed by running code.
llms.txt: One line per library, with where its documentation is.

A definition of done

The agent does not decide on its own that it has finished. One command runs the syntax checks, the static analysis and the tests of everything it changed, and the work is done when it passes:

php tools/mmo.php verify

The proof, measured

The same feature, an invoices page with a PDF export, built by the same agent on MMOFramework, on Laravel + Vue and on Symfony + Twig, five times each. Every run is measured from its transcript, and published with it.

	runs	done	tokens	minutes	errors fixed alone
MMOFramework	5	5	7 159 225	5.6	2
Laravel + Vue	5	5	2 396 438	5	0
Symfony + Twig	5	5	3 062 933	5.9	1

Medians of the runs, never one run.

Create a project live Read the documentation