mthds-plugins

The MTHDS plugin for Claude Code and Codex, which builds and runs AI methods with a runtime you run yourself.

For the easiest way to build and run methods from your coding agent, install the Pipelex plugin instead. It runs on the Pipelex hosted platform, which brings more features and a simpler setup: no runtime to install and no provider keys to manage.

What this repository is

This repository is the mthds-plugins marketplace. Its plugin, mthds, gives Claude Code and Codex the mthds-* skills and a hook that validates every .mthds file your agent edits. It needs no account and no hosted service: the skills drive the mthds-agent CLI, which runs your methods with the Pipelex runtime installed on your machine, on your own AI provider keys or a local model. With a Pipelex account, that same runtime can instead send your runs to the hosted Pipelex API, so that your machine holds no provider key; mthds-runner-setup sets up either way.

MTHDS is an open standard for AI methods, and a method runs on any MTHDS-compliant runtime; Pipelex is the reference one. Methods are shared on the MTHDS Hub.

Install

Paste the line for your agent into its chat, not into a terminal, and the agent runs the rest.

Claude Code

Install mthds: run bash -c 'npm install -g mthds && mthds-agent bootstrap && claude plugin marketplace add mthds-ai/mthds-plugins && claude plugin install mthds@mthds-plugins'. Then tell me to /reload-plugins.

Codex, 0.130 or later

Install mthds: run bash -c 'npm install -g mthds && mthds-agent bootstrap && mthds-agent codex apply-config && codex plugin marketplace add mthds-ai/mthds-plugins'. Then tell me to restart Codex and run /plugins to install mthds.

The manual commands, and what mthds-agent codex apply-config changes in ~/.codex/, are in Install the MTHDS plugin.

Skills

The skills work the same in both agents: /skill-name in Claude Code, $skill-name in Codex.

Skill	What it does
mthds-upgrade	Upgrade the MTHDS stack to its latest version
mthds-design	Design a new method bundle: contract first, written directly or step by step, validated until it runs
mthds-check	Validate a bundle, without changing it
mthds-edit	Modify an existing bundle
mthds-explain	Explain and document a method
mthds-fix	Fix validation errors automatically
mthds-run	Run a method and interpret its output
mthds-inputs	Prepare inputs: templates, synthetic data, files
mthds-install	Install method packages from GitHub or a local path
mthds-runner-setup	Set up how runs reach AI models: your own provider keys, or a Pipelex API key for the hosted Pipelex API
mthds-pkg	Manage MTHDS packages: init, dependencies, lock
mthds-publish	Publish methods to mthds.sh
mthds-share	Share methods on social media

Automatic validation

After every edit your agent makes to a .mthds file, the plugin's hook checks the file in three stages, and when the file is wrong it stops your agent and hands it the errors to fix:

1. Lint — plxt lint validates the TOML structure and the schema.
2. Format — plxt fmt formats the file.
3. Validate — the runtime's validator checks that the method is semantically correct.

A problem with your environment rather than with the file, such as a missing configuration, is passed to your agent as context without stopping it, and so is a formatting failure in Claude Code. A missing plxt stops the agent on every .mthds edit until it is installed. What each stage does in each agent is in Codex vs Claude Code hooks.

Documentation

• Install the MTHDS plugin: the manual install for each agent, and the Codex configuration.
• Codex vs Claude Code hooks: how the validation hook runs in each agent.
• The MTHDS standard: the language a method is written in.

Develop

The skills and hooks are rendered from the Jinja2 templates in templates/, once per build target; the plugin directories (mthds/, mthds-codex/ and the others) are generated, so edit the templates and rebuild:

make install   # create .venv and install the dependencies
make build     # render every target from templates/
make check     # the shared, Claude and Codex packaging checks
make test      # the unit tests

Build targets explains the targets and their variables.

License

MIT — Copyright (c) 2026 Evotis S.A.S.

Maintained by Pipelex. "Pipelex" is a trademark of Evotis S.A.S.