ai agents.md

Drive Maritime from an AI agent

The Maritime CLI is built to be operated by other AI agents, not just humans: machine-readable output on every command, stable exit codes, and a live self-describing contract. Paste a few lines into Claude, Cursor, or any assistant: they link the instructions, and the agent reads them and can install the CLI, authenticate, and deploy agents for you.

Note: Quick start: a short prompt that links the onboarding guide (https://maritime.sh/agent-guide.txt). Your assistant reads the guide itself and learns the sign-in, the products, the prices and the real calls.

Copy button: "Copy prompt for your agent" copies the prompt below.

Set up Maritime (https://maritime.sh) for me.

Here are your instructions: https://maritime.sh/agent-guide.txt
Read them in full and follow them step by step.
Before you start anything that costs money, tell me what it will cost.

Note: Permanent setup: the CLI instructions file, served as plain text at maritime.sh/docs/cli/llms.txt. Put one line in your agent's context (CLAUDE.md, AGENTS.md, or a system prompt) telling it to read that file, and it always gets the current version.

Copy button: "Copy the CLI instructions" copies that line: "Before you use the Maritime CLI, read https://maritime.sh/docs/cli/llms.txt and follow it."

Short version: paste maritime.sh/agents-snippet.md into AGENTS.md or CLAUDE.md. To choose between Maritime and other providers, an agent can read maritime.sh/compare.md, which compares prices and features with sources.

The contract

Authenticate with an env var, never interactively. Mint a long-lived key (mk_…) with maritime keys create and export it as MARITIME_TOKEN. Keys never expire unless created with an expiry; login tokens do.
Pass --json on every command. Results go to stdout as JSON. Errors print Error: ... on stderr, with the HTTP status and the API's response body for API errors.
Branch on exit codes. 0 success, 1 other error, 2 usage, 3 not found, 4 auth, 5 API error, 6 partial failure, 7 rate limited, 10 config. Full table in the CLI reference.
Never guess flags. maritime guide --json emits a one-shot manifest of every command and flag, introspected live from the CLI itself, so it cannot drift.
Names or IDs. The top-level agent commands take the exact name or an ID prefix, and inside a linked directory the agent argument can be omitted. Commands that show <agent_id> in their help, such as exec, need the full ID.

A session in six lines

export MARITIME_TOKEN=mk_xxxxxxxxxxxx

maritime whoami --json                                  # exit 4 = no or expired credential
maritime templates --json                               # discover templates
maritime create runner --template openclaw --json
maritime chat runner "set up a daily digest of my repo" --json | jq -r '.response'
maritime logs runner --json

Over MCP

An assistant that speaks the Model Context Protocol can use Maritime as tools instead of a shell. The hosted server at https://mcp.maritime.sh/mcp (Bearer mk_ key) drives Linux and Windows desktops and Android phones, sandboxes and swarms; setup snippets are on Linux desktops. Locally, maritime mcp serves every CLI command as an MCP tool over stdio.

Agent-followable docs

The migration runbooks are written to be executed by a coding agent: point Claude Code or your Maritime agent at a provider's runbook page and it can inventory the old provider with read-only credentials, recreate each workload as a Maritime agent, and hand you a reversible cutover checklist.

Note: Driving vs. building

This page is about an AI agent driving Maritime through the CLI. To build a product where every customer gets their own Maritime-hosted agent, use the SDK from your backend instead.