Which of your MCP servers is actually costing you tokens?
Paste your mcpServers config below and see an estimated token cost per server, worst first, with tool-search deferred loading on or off.
Those numbers are this project's own per-server catalog (tokens/session, tool-search deferred loading off), not a category average - see it price your own servers below. Other checkers in this space, like mcp-checkup on GitHub or policylayer.com's token-cost calculator, are worth comparing against; this one matches the exact server names in your mcpServers config - GitHub, Sentry, Supabase, and 7 more - against that curated catalog, prices anything else with a conservative generic profile, and ranks your real list worst offender first. The full context-budget picture is in the context window calculator.
A clean setup is just the beginning.
UsageCut trims that for free. Want Claude Code to be genuinely better, not just leaner? ClockedCode is the curated, conflict-free setup - installed in one paste.
Upgrade your Claude Code- 10x your Claude Code with hand-picked tools
- Specialist agents on your team
- Stop spending hours researching plugins
Priced per server, not guessed at in the abstract
Most writing on MCP token overhead describes the problem in general terms - servers are expensive, tool schemas add up - without ever pricing a real config. This checker runs your actual server list, whether pasted from ~/.claude.jsonor added by hand, through the same curated per-server catalog this project's own scanner uses: an instruction-block size and a full-schema size for well-known servers like GitHub, Playwright, and Supabase, and a conservative generic profile for anything it doesn't recognize.
The deferred-loading toggle mirrors a real Claude Code setting, not a hypothetical one - turning tool search on hides full schemas behind on-demand loading but still surfaces every tool name into context, so the checker shows both numbers instead of treating deferral as a free win. For the servers you're not sure you even use, see how to find and disable the MCP servers you're not actually using, or run the free scan below for the usage-based version of this same question. Server schemas are only one line item in what actually reaches the model each turn - see what a hook adds to that same bill for the rest of the picture.
More free tools for Claude Code
Small, focused, no-signup tools from the maker of ClockedCode. Free to use, forever.
More free Claude Code tools are on the way - built one at a time by a solo dev who ships daily.
Claude Code tips, every Sunday
One short email a week: the token-saving tricks, setup tweaks, and tools worth your time. Free, unsubscribe anytime.
About this estimate
Where do I get my mcpServers config to paste in?
It lives in ~/.claude.json (the top-level mcpServers key) for servers added globally, or in a project's .mcp.json for project-scoped ones. You can paste either the whole file or just the mcpServers object - both parse the same way.
Is this the exact token cost, or an estimate?
An estimate. It's built from the same curated per-server catalog this project's own scanner uses - instruction-block size and full-schema size for well-known servers like GitHub, Playwright, and Supabase - with a conservative generic profile for anything not in that list. Your real number depends on the server's actual current tool set, which only the free scan reads directly from your machine.
What does the tool-search / deferred loading toggle change?
Claude Code can load a server's full tool schemas upfront, or defer them and surface just tool names until one is actually called. Deferred loading still costs something - every tool name plus a small wrapper lands in context regardless - it just replaces the much larger full-schema cost with a smaller one. The toggle shows both numbers so you can see what turning it on is actually worth for your specific server mix.
A server here is cheap. Does that mean I should keep it?
Not necessarily - this only prices how big a server's schema is, not whether you ever call it. A five-tool server you use every session and a seventy-tool server you configured once and forgot about can cost about the same here while being completely different problems. Run the free scan for the second half of that picture: which servers your actual transcripts show going unused.
Questions, answered
Is UsageCut really free?
Yes. The scan and every fix are free - no card, no subscription, no signup. You only give an email if you want the optimization plan and one-command undo sent to your inbox.
Is my code safe? Does UsageCut upload anything?
The scan runs entirely on your machine. It never uploads your code, never reads your API key, and never sends your prompts or conversations anywhere. Only anonymous summary counts leave, and only if you choose to share them.
How does it reduce Claude Code token usage?
It reads your local setup and session history to find waste - idle MCP servers loaded into every session, an oversized CLAUDE.md re-sent on every request, duplicate file reads, and bloated tool output - then trims it losslessly, so each session carries less context and you hit your limit far less often.
What is ClockedCode?
ClockedCode is the curated, conflict-free Claude Code setup from the same maker. UsageCut makes your current setup leaner; ClockedCode makes it genuinely better - a set of hand-picked tools, specialist agents, and a tuned CLAUDE.md, installed in one paste.
Who is ClockedCode for?
Developers who use Claude Code daily and want a setup that is powerful out of the box instead of spending hours researching and wiring up plugins, agents, and MCP servers themselves.
What's included in ClockedCode?
A vetted, conflict-free bundle of hand-picked tools, specialist agents, and a tuned CLAUDE.md - everything installed together in one paste, with no setup archaeology.
Is ClockedCode a one-time payment?
Yes. ClockedCode is a one-time purchase with an instant download - no subscription and no recurring fees.
Does ClockedCode work with my existing setup?
Yes. It installs alongside what you already have and is fully reversible. Run UsageCut first to clean your current setup, then add ClockedCode to level it up.