Documentation

Connect the GoShip recorder to your AI coding tool and see what it records.

What GoShip does

GoShip connects a recorder to your AI coding tool through MCP (Model Context Protocol). The recorder logs your prompts, tool calls, which files changed and which commands passed. It never sends file contents. GoShip scores the session on six dimensions: AI Fluency, Output Quality, Judgment, Speed, Prompt Precision and Recovery Speed. The methodology page has the details.

IDE Setup

Pick your tool. Replace your-token with the token from your challenge setup page.

Claude Code

Run this in your terminal:

claude mcp add goship -e GOSHIP_TOKEN=your-token -- npx -y @goship/mcp-server

Cursor

Add to your project's .cursor/mcp.json:

{
  "mcpServers": {
    "goship": {
      "command": "npx",
      "args": ["-y", "@goship/mcp-server"],
      "env": {
        "GOSHIP_TOKEN": "your-token"
      }
    }
  }
}

Windsurf

Add to your project's .windsurf/mcp.json:

{
  "mcpServers": {
    "goship": {
      "command": "npx",
      "args": ["-y", "@goship/mcp-server"],
      "env": {
        "GOSHIP_TOKEN": "your-token"
      }
    }
  }
}

Privacy

Captured

Tool calls and MCP interactions
Files created, changed or deleted (paths and line counts)
Git commits with diff stats
Terminal commands and whether they passed
Your prompts, exactly as typed
How often and when you edit

Never captured

File contents
.env files and secrets
Keystrokes
Screen recordings
Files outside the project folder
API keys or credentials

Scoring

20%

AI Fluency

The context you give, the tools you call and the models you pick.

25%

Output Quality

Whether tests and builds pass, plus code quality, architecture and security.

20%

Judgment

What you do with each AI suggestion: accept it, edit it or override it.

15%

Speed

Time to a working result and iteration pace.

10%

Prompt Precision

Specific prompts that get it right without many retries.

10%

Recovery Speed

Time from a failing test or build to the next passing one.

Create an account and each challenge's setup page gives you the command with your token filled in.