Step 1
Connect your assistant once
Point your coding agent at letsship.ai/SKILL.md, or add https://mcp.letsship.ai/mcp to any client that supports remote MCP servers. Use a token bound to one project.
Connect Claude Code, Codex or any MCP client to SHIP once. Then ask it to send the work, and get back a reviewed, tested pull request.
Answer 4 questions to request access. SHIP is invite-only for now.
set up https://letsship.ai/SKILL.mdStep 1
Point your coding agent at letsship.ai/SKILL.md, or add https://mcp.letsship.ai/mcp to any client that supports remote MCP servers. Use a token bound to one project.
Step 2
Say what you want built. Your assistant files it as a mission with your notes, or hands over a plan it wrote or a pull request it opened.
Step 3
SHIP builds the change, waits for your CI, then reviews and tests it. Your assistant can read the verdict and the proof and answer with feedback.
Any client that accepts a remote server URL and an authorization header can file missions, follow them and send feedback. The server keeps no session, so a client can reconnect at any point.
Mint a token for the assistant with only the scopes it needs and bind it to one project, so everything it can reach stays inside that project.
Send feedback as an instruction and the Builder goes back to the open pull request with it. The reply says what the instruction will start.
Read the trace of a test session or a screenshot of the change working, attached to the mission as an artifact.
A mission runs for tens of minutes and carries on whether or not the assistant that filed it is still in the conversation.
Claude Code and Codex work through the SHIP skill and CLI, and any assistant that supports remote MCP servers works through https://mcp.letsship.ai/mcp.
No. Over MCP your assistant files missions against your project, and SHIP works on the repository in an isolated sandbox.
One SHIP credit covers one mission of up to 3 agent-hours, and each further 3 agent-hours, or part of them, uses one more credit. Credits come back when a mission ends in failure or needs attention with no pull request. Model usage is billed separately by your own provider.
Vibe-coded work is ready for production once it has passed your CI, a code review and a test of the running change. SHIP waits for your CI, reviews the change and tests it on whatever your assistant hands over, then leaves the pull request for you to merge.
Not by default. Every change arrives as a pull request for you to merge, and auto-merge stays off unless you turn it on for a project or a single mission.
Connect your assistant once and let every idea come back as a verified pull request.
Answer 4 questions to request access. SHIP is invite-only for now.