Use plone-client with coding agents#
This guide shows you how to let a coding agent read and change a Plone site through the command line, without giving it your passwords.
It works with any agent that can run a command.
An agent needs three things: the plone-client command on its path, profiles that you created, and a skill file in a directory it reads.
If your agent has no shell, connect it through the MCP server instead, as Connect an assistant through the MCP server shows. To follow all of this step by step on a throwaway site, take Your first agent session.
Create the profiles yourself#
Create the profiles in your own terminal, so the secrets go into your keyring and never through the agent. Make every site other than your local one read-only.
plone-client profile add blog-local --url http://localhost:8080/Plone --username admin --frontend-url http://localhost:3000
plone-client profile add blog --url https://example.org --username editor --read-only
Point the project at its profile#
Add a .plone-client.toml file at the root of the project.
profile = "blog-local"
Install the agent skill#
plone-client ships an Agent Skill that teaches an agent its commands, output, exit codes, the Volto content model, and the safety rules: no passwords, a dry run before changes, and no deletion without your confirmation. Agent skill shows its full text.
Install it in the directory your agent reads, at the root of the project.
plone-client skill install
This writes .claude/skills/plone-client/SKILL.md.
plone-client skill install --dir .agents/skills
plone-client skill install --dir .agents/skills
Cursor also reads .cursor/skills, and .claude/skills for compatibility.
plone-client skill install --dir .agents/skills
GitHub Copilot in VS Code also reads .github/skills and .claude/skills.
plone-client skill install --dir .agents/skills
Gemini CLI also reads .gemini/skills.
Commit the file to share it with everyone working on the project.
To install it for every project instead, pass the directory in your home. Agent skill lists these per agent.
plone-client skill install --user
plone-client skill install --dir ~/.agents/skills
--user writes to ~/.claude/skills.
Run the command again after upgrading plone-client, to replace the skill with the new version.
Use an agent without skill support#
Print the skill, and paste it into the agent’s rules or its first message.
plone-client skill show
Tell the agent about your sites#
The skill knows the command line, not your sites.
Add what only you know to the file your agent reads for the project, such as AGENTS.md, CLAUDE.md, or GEMINI.md.
Start from this template.
# Plone sites
## Profiles
- `blog-local` is the local site and the default. You may change it.
- `blog` is production, and it is read only. Pass `-p blog` to read it.
- If a command exits with code 7, tell me. Do not work around it.
## Content
- News goes in `/blog`, as `News Item`, published with the `publish` transition.
- Documentation goes in `/docs`, as `Document`.
## Rules
- Never ask me for a password, and never call the REST API with `curl`.
- Never create or remove profiles unless I ask.
- Run a change with `-n` first, and show me the output.
- Delete nothing unless I confirmed the exact items.