Run Claude Code on MIOSA

What you're building

Anthropic's terminal coding agent Modelled on Claude Code, by Anthropic.

Claude Code reads a repository, edits files, runs commands and tests, and works in a loop with the user or on its own; it supports tool servers (MCP) and sub-agents.

Primitives you'll use: Sandbox · Agent and harness

• Understands and edits a whole repository
• Runs commands and tests
• Works in the terminal or headless
• Extensible with MCP tools

You install and run the real Claude Code on MIOSA. Read the official docs first: code.claude.com/docs/en/setup. Every install command below comes from them; if the docs and this guide disagree, the docs win. Claude Code is Anthropic's CLI; installs with an official install script.

What you need on MIOSA

Each row is one thing to create before you start. The number matches the step that uses it.

1. 1 Organization and API key Scopes every call; a workspace key is all a worker needs. miosa api-key create app-key --preset agent Docs
2. 2 Sandbox The isolated Linux workspace the agent writes code and runs commands in. miosa create app-box --template nextjs --wait Docs
3. 2 A workspace per customer Isolates each customer’s machines, deployments, and data as they onboard. miosa workspace create customer-1 Docs
4. 2 Branding and white-label Your name and slug on previews, deployments, and the desktop; customers never see MIOSA. miosa org Docs
5. 2 Usage metering and bill-to Usage per customer, and which account pays for new machines. miosa org bill Docs
6. 3 Agent and harness Turns a prompt into work: pick the harness and model a run uses. miosa agent harnesses Docs

Architecture

How it runs on MIOSA

Where it runs -> a MIOSA sandbox (an isolated Linux workspace)

State and files -> Keep the repo and the `~/.claude` config on the persistent sandbox filesystem; both survive pauses.

Model -> your own provider key, never a MIOSA platform key

MIOSA can also dispatch it as the built-in `claude-code` harness: runs, events, files and approvals come with it

Your customers -> one workspace each, isolated from each other

Auth and tenancy

Your organization is the platform; customers never get a MIOSA account, they sign into YOUR product. One workspace per customer, created as they onboard, isolates their machines, runs and data.

miosa workspace create claude-code-customer-1

Tag every machine and run with the customer id in `metadata`, and never query across customers. Meter usage per customer with GET /api/v1/usage and set who pays with PUT /api/v1/bill-to (/docs/platform/usage-and-billing).

Install Claude Code

Run these inside the sandbox (for example `miosa exec claude-code-task-1 -- bash -lc '<command>'`), in order. The commands are from the official docs.

curl -fsSL https://claude.ai/install.sh | bash

Check: the install finishes without errors; the next sections configure and start it.

Configure

Authenticate with your own Anthropic key: export ANTHROPIC_API_KEY="..." (or run `claude` once and sign in). Never use a platform key.

Model: Anthropic (Claude). Calls use my own provider key (ANTHROPIC_API_KEY); MIOSA platform keys are never used.

export ANTHROPIC_API_KEY="..."   # set it in the process environment, or the Secrets API; never write it into files you commit

Run it

Start it and prove it works.

claude --version

Scripted, non-interactive use (what a product or a queue would call):

claude -p "summarise this repository and list the failing tests"

MIOSA also runs it natively. Dispatch a task and stream its events, files and approvals through MIOSA:

miosa prompt --sandbox claude-code-task-1 --harness claude-code --model <anthropic-model-id> "summarise this repository"

Sessions: one chat per project or conversation, so the agent keeps its context. The first `miosa prompt` on a sandbox uses `--new-chat` (a chat id is printed); every later turn passes `--chat <chat-id>`. `--reuse chat` keeps one new machine per chat so files persist.

miosa run follow <run-id>

Check: a headless task completes and prints a result.

Persistence and access

Keep the repo and the `~/.claude` config on the persistent sandbox filesystem; both survive pauses.

Check: after the machine pauses and resumes, the tool starts again with its state intact.

Make it a product

Run each customer task in that customer's workspace with `miosa prompt --harness claude-code` and stream the events into your own UI.

Limits and costs

Pin the version you tested and update deliberately; these tools change quickly.

Give the tool only the credentials it needs, as secrets, not baked into files.

Prototype: keep it to one machine at the default size, skip replicas and custom hostnames you do not need, and delete everything when you are done.

Acceptance checks

Claude Code runs on a MIOSA sandbox and answers a real task.

State survives a pause and resume.

Each customer has its own workspace and its own instance; nothing is shared between them.

Everything it created can be deleted with nothing left running.

Next steps

• Agent prompt builder Pick a product shape and hand the brief to your agent.
• Run coding agents Run Claude Code, Codex or Cursor in the cloud, or build a product on them.
• Compare MIOSA How MIOSA compares to other sandbox and agent platforms.
• What you can build Every product pattern MIOSA supports.