Why you'd want mods in Claude Code (and which kind to install first)

You do not need mods to use Claude Code. Here is what changes when you add them, sorted by how quickly you will feel it.

2026-09-16 · Claude Mods editorial team

Claude Code works fine without mods. So why are dozens of people writing them in the first two weeks of early access? Because a terminal agent leaves a lot of questions unanswered while it works, and a mod can make useful information visible inside the session. Here are the five things people install mods for, in the order you are likely to notice them.

1. You can finally see what is going on

The first category of mods to appear was dashboards. How full is the context window? How much has this session cost? When does the plan’s five-hour window reset? Claude Code has some of this behind slash commands; a mod pins it under the prompt and keeps it live.

  • context-lens shows the context-window fill, how fast it grows per turn and how many turns are left before compaction.
  • token-ledger keeps a running cost and cache-hit ratio on one line.
  • cctop is the maximalist version: a btop-style panel with tokens, cost, rate limits, latency per tool and subagents.
  • agent-flow and flowpane draw the tree or graph of subagents when a task fans out, so you can see who is stuck.

None of this touches the model. The numbers come from the session object a mod can read.

2. Guardrails that actually hold

A rule in a prompt is a request. A mod that denies a tool call is a fact. That difference is why safety mods exist.

  • budget-guard warns near a cost limit and then refuses tool calls until you confirm.
  • secret-redactor swaps secrets, emails and IPs for placeholders before the model reads them, and restores them on the way into the tool. The model never sees the real value.
  • Anthropic’s own sec-default sits outermost on managed machines so that nothing a user installs can wrap the organisation’s policy.

Classic shell hooks can block a call too, but they see the command as a string and have to parse it. A mod gets the tool name and its arguments as typed data, which can make event handling more precise. The actual protection depends on its implementation and allowed capabilities.

3. The wait

An agent that takes ninety seconds to answer leaves you staring at a spinner. The “while you wait” mods are the most visible thing in the ecosystem and, honestly, the reason many people enable the flag.

  • cc-arcade puts nine games above the prompt and pauses them when the answer lands.
  • Mindful Claude turns the wait into a breathing exercise; the spinner counts the breath with you.
  • Storytime runs a tiny language model inside the mod and narrates your session as a short story.
  • Claude Slots replaces the spinner with a slot machine; cc-maid gives you a pixel companion whose expression follows Claude’s mood.

Local rendering does not itself require model tokens; normal Claude usage and any model calls a mod makes still have their usual costs. Whatever you think of playing Tetris in a terminal, the mechanism is the same one the serious mods use.

4. Glue for your workflow

Some of the best mods are small and boring in the best way: they pull one thing you keep alt-tabbing for into the terminal.

  • cc-pr-tracker pins the pull requests you watch, with merge state and required checks, and toasts when one changes.
  • gh-ci-status does the same for GitHub Actions runs of the current repo.
  • claude-queue lets you type the next instruction while the current one runs, and sends it when the turn ends.
  • aside opens a read-only side chat about the session so far, without disturbing the main thread.

5. Changing what Claude Code shows you

The last group is the one that would have been impossible before: mods that change rendering.

  • claude-mermaid draws every mermaid block Claude writes as box art in the transcript.
  • time stamps each of your messages with the time you sent it.
  • cc-mod-waitwhat retells a long answer in plain words in a band above the prompt.

Anthropic’s demo of the same idea hid sensitive values in the desktop app until you hover, so you can share your screen. Expect more of these once the API settles.

What to install first

If you only try one: cc-arcade for the demo effect, or context-lens if you want something useful on day one. Turn the flag on once (see the install guide), install with two commands, and you can vote for the mod of the day once you have an opinion.

One caution. Mods run inside Claude Code with the permissions of your session. Read the code of anything you install, prefer mods whose listing says what they hook and what they can reach, and treat “community listed” entries with more care than “verified” ones.

Frequently asked questions

Which mod should I try first?

Choose one that solves a specific need: a dashboard for session visibility, a workflow tool for repeated tasks, or a game for waiting time. Start with one and check compatibility.

Can I use Claude Code without mods?

Yes. Mods are optional extensions. You do not need them for normal Claude Code use.

More reading