Stretch your Claude Code plan.
Every Claude Code session starts with ~100k tokens of context before you type, re-sent on every turn and to every subagent. limitcast measures exactly what's in it, shows what you never use, and stops runaway subagents before they burn the rest of your window. More work out of the plan you already pay for.
$ limitcast trim Every Claude Code session starts at 102.7k tokens before your first message. ## Measured savings: ~6.2k tokens per session (6% of your starting context) - Disable 9 MCP servers you haven't used in 14 days (~0.9k): postman, context7, ... - Hide 66 skills from the model (~5.3k): dataviz, block-kit, ... | Instruction files 8.2k | Tool names 7.1k | | Skill listing 6.7k | SessionStart hooks 6.5k | | Agent list 4.9k | MCP instructions 3.1k |
A real run on a heavy user's setup, names shortened. Your numbers come from your own transcripts.
How it works
Trim advisor (Pro)
Measures what your latest session loaded before you typed: hook output, MCP servers, tool names, skills, agents, instruction files. Then names the ones you haven't used, with the tokens each one costs you.
Subagent guard (Pro)
Blocks a new subagent in Claude Code or Codex when that window is on pace to run out before it resets. Never blocks once you're past 100%, so extra usage keeps working.
Burn report (Pro)
Claude Code, Codex and Grok tokens by day, project, model and session, plus the share your subagents use.
Runout forecast (free)
Burn rate and the clock time each 5-hour and weekly window runs out, for Claude Code and Codex.
How it compares
| limitcast | ClaudeBar | ccusage | |
|---|---|---|---|
| Price | Free core, Pro $79/yr | Free | Free |
| Usage and quota display | Claude Code, Codex, Grok | 10+ tools, menu bar | 18 tools, token and cost reports |
| When you'll run out | Clock time from your burn rate | Pace label (running hot / on track) | 5-hour block tracking |
| Stops subagents before you run out | Yes, Claude Code and Codex | No (hooks observe only) | No |
| Measures what loads into every session, and what to cut | Yes | No | No |
| Uses your Claude login token | No | Yes, for its usage API probe | No |
| Runs on | macOS, Linux, over SSH | macOS | Command line (npx / bunx) |
From each project's public README and source, October 2026. If you only want to see your numbers, the free tools are great.
Install in 30 seconds
npm install -g limitcast limitcast install # wraps your status line, backs up settings.json limitcast install --guard # Pro: also adds the subagent guard to Claude Code and Codex limitcast doctor
Requires Node 20+, macOS or Linux. limitcast uninstall restores your original settings.
Pricing
Pro
- Everything in Free
- Trim advisor: measured savings per session
- Subagent guard for Claude Code and Codex
- Burn report across Claude Code, Codex and Grok
- Email support
FAQ
Does it send my code or prompts anywhere?
No. limitcast reads the limit readings Claude Code passes to your status line and the logs Claude Code and Codex already keep on your disk. Nothing leaves your machine, and license keys are checked offline.
Does it use my Claude login?
No. It never touches your subscription credentials, so it can't put your account at risk.
I use extra usage past my limit. Will the guard block me?
No. Once a window is used up, limitcast steps aside and leaves it to Claude Code. The guard only acts while you're heading toward a limit.
Does it work with Codex and Grok?
Codex: yes, forecasts, the subagent guard and the burn report, from the logs Codex already writes. Grok: the burn report covers your Grok tokens, but Grok keeps no limit readings on disk, so there's no Grok runout forecast and the guard never blocks Grok.
What if I have two Claude accounts?
Each session is forecast from its own readings, so parallel accounts never mix.
Can I get a refund?
Yes, within 14 days, no questions asked.