Glossary

.mcp.json is a small file kept in your project that lists which MCP connectors should be available there. When the AI starts, it reads this file to know which outside tools to switch on, like a web search or a database link. Sharing your .mcp.json lets someone else load the exact same connectors with no guesswork. The .json ending just means it is a structured text file.

An agent is an AI helper that takes a goal you describe in plain words and works toward it on its own. Instead of you running every step by hand, the agent decides what to do next, uses the tools it has, and reports back. In Setuproll, most of the things people share are agents they have shaped for a specific job.

A build is a finished, shareable package of how someone set up their AI helper for a certain kind of work. It bundles the instructions, the tools, and the small choices that make it run well, so you can copy the whole thing instead of starting from zero. Think of it like a recipe card: follow it and you get the same result the author got.

CLAUDE.md is a plain text file you keep in your project that the AI reads first, every time. You write down the things you always want it to know, such as how you like your code formatted or which folders to avoid. It saves you from repeating the same reminders in every chat. The name and the .md ending just mean it is a simple text file.

Config, short for configuration, is the set of saved choices that tell a tool how to behave. It covers things like which options are turned on, which model to use, and where files live. Once you write a config down, the tool reads it instead of asking you the same questions again. Sharing a good config is one of the easiest ways to help someone copy your setup.

The context window is how much text the AI can hold in mind at one time, including your messages, its replies, and any files it has read. Picture it as a desk with limited space: once the desk is full, older notes have to be cleared to make room for new ones. A bigger context window lets the AI keep track of more before it starts losing the earlier details. This is why very long sessions sometimes seem to forget what was said at the start.

Hooks are small actions you set up once so they run automatically at a chosen moment, such as right before or right after the AI does something. For example, a hook could run a formatter every time a file is saved, or send you a note when a task finishes. You decide the trigger and the action, and the tool handles the rest. They let you bake your habits into the workflow so you never forget a step.

An MCP server is a connector that lets the AI reach an outside tool or service in a safe, agreed way. With one running, the AI can do things like read a database, search the web, or pull files from a drive, all without you copying and pasting. MCP stands for Model Context Protocol, which is just the shared rulebook these connectors follow. You add the ones you need and skip the rest.

A model is the underlying AI that actually reads your words and produces a reply. Different models have different strengths: some are faster, some are cheaper, and some are better at hard reasoning. Choosing a model is like picking the right engine for the job. Many builds note which model they were tuned for, because the same setup can feel different on another one.

Pass rate is the share of checks a build got right, written as a percentage. If a build is run against 100 small tests and gets 87 correct, its pass rate is 87 percent. A higher pass rate suggests the setup is more reliable at the kind of task being measured. It is one quick way to compare builds without trying each one yourself.

A score is a single number that sums up how well a build did across the things Setuproll measures. It rolls together signals like accuracy, speed, and reliability into one figure so you can compare at a glance. A higher score means a stronger overall result. Use it as a starting point, then read the details to see if a build fits your own needs.

settings.json is a file that holds your preferences for the tool, such as what it is allowed to do, which features are on, and any shortcuts you set. The tool reads it on startup so your choices stick from one session to the next. Editing this file is how you fine tune behavior beyond the basic menus. The .json ending means it is a structured text file you can open and read.

Setup is everything you put in place to get a tool ready before you start real work. It includes installing what you need, signing in, picking your options, and adding any files the tool expects. A good setup means you can begin without hitting surprises later. People often share their setup steps so others can reach the same starting point quickly.

A slash command is a shortcut you type that begins with a slash, like /review or /test, to trigger a saved action right away. Instead of writing out the same long request each time, you type the short command and the tool runs the steps behind it. You can use built-in ones or make your own for tasks you repeat. They keep common jobs fast and consistent.

A subagent is a smaller helper that the main AI hands a focused side task to, so it can keep its own attention on the bigger goal. The subagent works on its piece, then passes the result back. This is handy for jobs like searching many files or running a long check, because it keeps the main conversation clear. You can think of it as bringing in a specialist for one part of the work.

A tier is a rank that sorts builds into quality groups, usually from S at the top down through A, B, C, and D. It gives you a fast sense of where a build stands without reading every number. Tiers come from the underlying scores, then get grouped so similar results sit together. They are a friendly way to browse when you just want the strong options.