SHIP

What SHIP MCP server does

The SHIP MCP server turns tracked software work into managed missions for an AI coding agent team. A mission can begin from an existing Linear, Jira, or GitHub issue, or the server can create the issue as part of starting the work. The workflow assigns separate stages to a Planner, Builder, Reviewer, and Tester, with CI acting as a gate between stages.

Agents can produce code changes and open a pull request in the connected GitHub repository. From an MCP client, you can follow the mission, provide additional instructions, pause or stop it, approve continuation after a credit limit, and decide whether to merge or decline the pull request.

How it works

The service is available at https://mcp.letsship.ai/mcp over Streamable HTTP. It is stateless, so a client can reconnect without losing mission state. A mission normally starts with planning, but an existing plan can send it directly to the Builder. An open pull request can route the work to review or testing instead.

Mission records include stages, verdicts, provenance, session logs, pull requests, agent runs, and test or review results. Artifacts such as test traces and screenshots can be listed and opened through dedicated tools. The roles can use a selected harness and model, including Claude Code, Codex, OpenCode, Kilo Code, or the pi coding agent, according to the project configuration.

Setup and configuration

Add the remote endpoint to a compatible MCP client. OAuth 2.1 clients can discover the authorization flow and complete sign-in through the SHIP console. Clients without OAuth can send a personal access token as a Bearer token; tokens are created in SHIP settings and can be restricted to one project.

Claude Code, Cursor, VS Code, and Codex each have documented HTTP configuration formats. A client that supports only stdio can connect through mcp-remote. The supplied configuration can also include an Authorization header for token-based access.

Project tools expose reachable projects, project configuration, connections, auto-merge settings, issue codes, and default sandboxes. Scheduled-task tools can create, edit, disable, run, and inspect recurring missions and their past runs.

Tools and capabilities

The SHIP MCP server groups its tools around mission control, project administration, proof, and scheduling:

• Start work from an existing issue or create a new issue and mission.
• Send instructions to active missions and record notes for later stages.
• List, inspect, pause, resume, stop, archive, and unarchive missions.
• Read reviews, tests, events, agent runs, pull requests, and run artifacts.
• Override a role’s harness, model, prompt, or checkpoint for one mission.
• Set mission-specific auto-merge behavior and record an ending disposition.
• Merge or decline a mission’s pull request.
• Create and run scheduled missions, then review their costs and results.

Starting a mission or running a scheduled task consumes one credit. Messages sent to an active mission do not consume credits.

Limitations and notes

The server operates against connected external systems, so token scopes determine which projects and actions are available. Destructive or externally affecting operations include stopping missions, merging or declining pull requests, changing project settings, and approving additional credit use. Review client confirmation behavior before enabling such actions.

The README documents the remote service and client configuration rather than a local package installation. A personal access token must be supplied explicitly when OAuth is unavailable. The pull request remains open when a mission is stopped, while declining closes it without merging.