Integrate

Use with AI agents

mountOS ships a machine-readable skill layer. Point an AI agent at it. The agent can then set up, operate, and use a deployment completely.

The skill index routes an agent to the one focused skill that matches the task. The corpus files publish the whole site as text for agents that ingest rather than browse. You can copy any docs page directly into an assistant.

Point your agent at the skill index

Give a Claude agent, or any MCP-capable client, one line. It reads the index. It then follows the links from there.

Read https://mountos.io/skill.md

/skill.md is the entry point. It carries the mental model that an agent must not violate. Then it routes to the focused skills below.

Installable skills

The skill index above is enough for a question. For real work, install the mountOS agent skills. They are open source, and they add what reference documentation cannot: the order of operations, the check that proves each stage, and the failure modes that report healthy while the system is broken.

Skill	For	What it adds
`deploy`	Operator	Stand up, verify, and operate a deployment. Architecture, the ordered bring-up, per-stage verification, and the failure modes that report healthy while the system is broken.
`integrate`	Application developer	Add mountOS under a product that already has customers. Mapping an existing user base, credential shapes, and picking a data surface per workload.
`troubleshoot`	Whoever is on call	Diagnose a deployment that is failing. What evidence to gather in what order, and which checks prove nothing.

They are independent, so install only the ones you need. Each is plain Markdown and names no specific agent. Pick the route that matches your tool.

Any agent, no install

Give it one line. This works with any agent that can fetch a URL or take a pasted document.

Read https://raw.githubusercontent.com/mountos-io/skills/main/deploy/deploy.bundle.md and follow it.

That file is one whole skill in a single document, entry point plus every reference, with the cross-links rewritten as in-document anchors. Swap deploy for integrate or troubleshoot to load a different one. Paste it into a system prompt if your tool cannot fetch.

Clone, for the multi-file version

git clone https://github.com/mountos-io/skills.git ~/.mountos-skills

The multi-file form loads references on demand instead of all at once, so it costs less context. Agents that read AGENTS.md pick it up on their own once the repository is in the workspace. For Claude Code, link it:

ln -s ~/.mountos-skills/deploy ~/.claude/skills/deploy
ln -s ~/.mountos-skills/integrate ~/.claude/skills/integrate
ln -s ~/.mountos-skills/troubleshoot ~/.claude/skills/troubleshoot

Link only the ones you want. Each is then available by its name, for example /deploy, and the agent loads it on its own when a task matches. For any other tool, copy the skill directory into whatever skill or rules directory it uses.

Claude Code plugin

The repository is also a plugin marketplace, which handles install and updates for you.

/plugin marketplace add mountos-io/skills
/plugin install deploy@mountos

Each skill is a separate plugin, so integrate@mountos and troubleshoot@mountos install the same way, independently.

Staying current

The skill loads /skill.md and /llms.txt on every use. Facts that change with a mountOS release therefore stay correct with no skill update. The guidance in the skill is versioned on its own. Check for a newer release:

curl -fsSL https://raw.githubusercontent.com/mountos-io/skills/main/deploy/VERSION

Then upgrade in place:

git -C ~/.mountos-skills pull --ff-only

Focused skills

Each skill is a single Markdown file. It covers one job. An agent reads the index. Then it loads only the skill that the task needs.

Skill	What it covers
`provision`	Set up a deployment completely: the HUB, account, region, and services.
`volumes`	Create and use volumes: storage, keys, mounts, forks, and time travel.
`integrate`	Connect over S3, WebHDFS, Kubernetes CSI, the change-event feed, and the Admin SDK.
`operate`	Run day 2 operations: scale metadata clusters and pools, move volumes, rotate keys, and run lifecycle tasks.
`s3`	Tune S3 clients and multipart uploads for AWS CLI, boto3, rclone, DuckDB, Trino, and PyIceberg.
`iceberg`	Query an iceberg lake volume from DuckDB, Trino, Spark, or PyIceberg over the REST catalog.
`deploy`	Provision the cloud substrate. Start the HUB on AWS, GCP, or Azure.
`env`	Look up each service's environment variables and their meaning.

Machine-readable corpus

mountOS publishes the site as text for agents that load a whole corpus instead of following links. /llms.txt is the topic index. It has one entry per page. /ai/llms-full.txt combines every docs page into a single file.

Copy any page to an assistant

Every docs page carries a Copy page control in the right rail. It copies the page as Markdown. You can paste the copy into an assistant. The same menu opens the raw Markdown. It can also hand the page to ChatGPT or Claude in one click.