Go from install to a useful answer.

Quickstart

From nothing installed to one answer you can check.

Steps two through five were run end to end on the exact two-file repository step two creates, using the npm-published release in an isolated scratch environment. Follow them in order; each step says what you should see before you move to the next one.

1. 01

   Install Kin

   One command, the same on macOS, Linux, Windows and WSL. It installs Kin and registers it with the AI coding tools on the machine. It needs Node.js 20 or newer; the install page carries a shell installer and a PowerShell one for a machine without it. Kin lands in ~/.kin/bin and the install adds that directory to your shell profile, so start a new login shell before the bare name works.

   npx -y @kinlab/kin setup
   exec "$SHELL" -l
   kin --version

   Copy command

   Prints the version you just installed, for example kin 0.7.19.

2. 02

   Prepare a small disposable repository

   Use a repository you do not mind experimenting on. This block refuses to touch an existing kin-quickstart directory, then creates two small files so a mistake costs nothing and you can check every answer below against source you can see right here.

   Create the disposable example: commands and source

   if [ -e kin-quickstart ]; then
     printf '%s
   ' 'Refusing to overwrite existing kin-quickstart directory.' >&2
     false
   else
     mkdir kin-quickstart &&
     cd kin-quickstart &&
     mkdir src &&
     git init -q -b main &&
     cat > src/greeting.js <<'EOF' &&
   function formatGreeting(name) {
     return `Hello, ${name}!`;
   }
   
   module.exports = { formatGreeting };
   EOF
     cat > src/cli.js <<'EOF' &&
   const { formatGreeting } = require('./greeting');
   
   function main(name) {
     console.log(formatGreeting(name));
   }
   
   main(process.argv[2] || 'world');
   EOF
     git add -A &&
     git commit -q -s -m "Add greeting helper and CLI entry point" &&
     kin init .
   fi

   Copy command

   kin init reports what it found (4 entities and 5 relations across the 2 files here) and starts a daemon for this repository.

3. 03

   Check graph readiness

   Do this before you trust an answer, not after.

   kin graph status

   Copy command

   Reports entity and relation counts, embedding coverage, and parse coverage. On a repository this small you may see a note such as "all entities are Source, role classification may not be working". That means a heuristic had too little code to calibrate against, not that anything failed. What matters here: embeddings report fully indexed, and both files show up under parse coverage.

4. 04

   Get one answer, and check it against the source

   Ask what calls the helper function, then open the two files above and read the same thing yourself.

   kin refs formatGreeting --kind calls

   Copy command

   Compare with the saved output

   References to 'formatGreeting' -> formatGreeting (Function) @ src/greeting.js
   referenced by 1 entities:
     main @ src/cli.js:3 [Calls] (type_resolved) sites 4

   main in src/cli.js calls formatGreeting in src/greeting.js. Open both files above: that is exactly what they do, so you have checked the answer rather than taken it on faith.

5. 05

   Connect an agent (MCP)

   Kin exposes the same graph to the AI client you already use.

   kin setup --intent agent

   Copy command

   Configures the MCP entry for every AI client it detects (Claude Code, Cursor, Codex CLI, Gemini CLI, and others) and prints a first prompt to try. Restart the client so it reads the new configuration.