AI Quickstart

Set up your coding agent with Sail’s docs MCP and workflow skills for migration and agent building.

Sail works with the coding agent you already use. Claude Code and local Codex sessions can install the full Sail plugin, including skills and delegation tools. Other tools can still use the documentation MCP server.

Overview

Start using Sail by providing your agent with Sail’s docs and the patterns to migrate existing apps or build observable agents on Sail.

Sail docs MCP

Connects your agent to Sail’s documentation: the API, models, pricing, and SDK. Served live from these docs, so it stays current.

Sail workflow skills

Gives your agent the patterns to migrate existing apps and to build, instrument, and debug observable agents on Sail.

Step 1: Give your agent Sail’s docs

Connect your tool to the Sail docs MCP server at https://docs.sailresearch.com/mcp. It gives your agent a live connection to these docs, so answers stay current:

• Claude Code: claude mcp add --transport http sail-docs https://docs.sailresearch.com/mcp
• Codex: codex mcp add sail-docs --url https://docs.sailresearch.com/mcp
• Claude Desktop: Settings → Connectors → Add custom connector, then paste the URL.
• ChatGPT: Where your workspace allows Developer mode and custom connectors, go to Settings → Apps & Connectors → Developer mode → Add new connector, then paste the URL.
• Cursor: Settings → MCP, then add the URL.

See the MCP server guide for more clients and example questions.

Prefer not to install anything? Every docs page has a copy button at the top if you want to paste content directly into your agent interface. Or point your tool at llms.txt or llms-full.txt.

Step 2: Give your agent Sail’s workflow skills

The canonical package is sailresearchco/sail-skills. It packages skills for delegating coding work, migrating existing apps to Sail, and building observable agents. An agent run is recorded as a Voyage: a trace of named agents, spans, and events, with every model call and Sailbox command attributed to the right step.

Skill	Use it when
sail-subs	Delegate scoped coding or analysis while the host agent keeps ownership of the larger task.
sail-swarm	Run an announced, host-owned campaign of coordinated workers over a shared surface.
sail-review	Request an on-demand, read-only code review with severity-ordered findings.
sail-migrate	Migrate an existing app, agent, or workflow to Sail while preserving behavior.
sail-voyage	Build or instrument a Voyage with agents, spans, events, Sailbox commands, and terminal lifecycle.
sail-inference-with-voyage	Attribute Sail inference model calls to the active agent and span.
sail-voyage-debugging	Diagnose a Voyage that ran but appears incomplete or incorrect in the dashboard.
sail-gpu-marketplace	Allocate and use preemptible GPU compute or recover checkpointed work after an interruption.

Install for your agent:

Claude Code

/plugin marketplace add sailresearchco/sail-skills
/plugin install sail@sail

The full plugin, including the sail-delegate MCP server, works in Claude Code and local Codex app, CLI, and IDE sessions. Hosted Codex sessions cannot run the local server. See Sail for coding agents for the three delegation modes, direct invocation syntax, and troubleshooting.

Before you run generated SDK code, install sail. The Sail SDK can use a credential stored by sail auth login on the same machine. For CI, deployments, Sailbox guests, or third-party inference SDKs configured for Sail, provide SAIL_API_KEY in that runtime environment.

Next steps

Already have an agent or workflow? Install the skills above, then ask your coding agent to “Migrate this app to Sail”. The sail-migrate skill guides the whole migration, including moving sandboxed execution to Sail. Prefer not to install anything? Copy the migration prompt from Migrate to Sail instead.

Starting fresh? Use the sail-voyage skill to build your own observable AI workflow. Ask your coding agent:

Build a small background research agent on Sail and make the whole run show
up as a trace I can open in the dashboard, each step, the model calls it
makes, and the sandbox commands it runs, attributed to the right part of the
workflow.

The sail-voyage skill takes it from there.

Or make a plain inference call first. See the inference Quickstart.