Ever catch yourself wondering if Claude’s secretly tallying your lazy days?
That’s the hook with cc-mcp, this MCP server that hands Claude real-time access to your Claude Code usage stats. No more firing up terminal tools, squinting at numbers, then forgetting them five minutes later. Now, mid-conversation, you ask: “How much Claude Code have I burned this month?” And bam — Claude spits back precise figures, backed by your local data.
“You’ve logged 47.3h interactive + 83.1h AI sub-agent work in March, for 130.4h total. You’re on a 36-day streak with 22 Ghost Days. Based on your last 14 days, your streak is likely to survive — you’ve been active 100% of days this month.”
Look, I’ve seen a lot of “analytics” tools in 20 years chasing Silicon Valley’s next big metric. Most are forgettable dashboards that collect dust. But this? It’s different. cc-mcp pipes that data straight into Claude’s brain during your session. Planning a marathon coding sprint? Claude factors in your 8 hours already logged today, the 3 hours of rogue AI sub-agents chugging away while you slept — yeah, those Ghost Days — and warns if your precious streak’s on the line.
What the Hell is an MCP Server, Anyway?
MCP stands for Model Context Protocol, some fresh protocol letting AIs call local tools like subprocesses. Think of it as giving Claude a Swiss Army knife for your machine — but only the blades you install. Here, cc-mcp spins up a Node.js server using @modelcontextprotocol/sdk. Each tool? Just wraps cc-agent-load –json, parses it, and serves structured JSON plus plain text.
Add it to your claude_desktop_config.json like this:
{ “mcpServers”: { “cc-toolkit”: { “command”: “npx”, “args”: [“@yurukusa/cc-mcp”] } } }
Restart. Boom. Four new toys: today’s totals, 7-day/month breakdowns, streaks, autonomy ratios. Per-project splits for your top 15 time sinks. Even month-end projections. “Will I smash 700 hours?” Claude crunches the math on the fly.
And Ghost Days — that’s the gem. Days when AI agents ran wild without you touching a keyboard. Useful? Hell yes, for spotting how much your workflow’s gone autonomous. Claude surfaces them naturally: “Your Ghost Days last week? Tuesdays and Fridays — AI kept the lights on.”
But here’s my unique spin, one you won’t find in the GitHub README: this echoes the early days of Unix process monitors like top or sar, where feedback loops turned sysadmins into power users. Back then, seeing CPU hogs in real-time changed how we coded. Now? cc-mcp closes that loop for AI devs. Bold prediction: in six months, expect every major LLM to demand its own usage telemetry — not for billing (yet), but for self-optimizing agent swarms. Who makes money? Not Anthropic directly — this is pure open-source grit from yurukura. But it juices their retention, keeps you hooked longer. Skeptical me asks: is this empowerment, or just gamified dependency?
Installation’s dead simple. npm install -g cc-agent-load. npx @yurukusa/cc-mcp to test. Config tweak, restart Claude Desktop. Zero telemetry, all local — data stays put, no phoning home. Works with any MCP client, part of a 27-tool cc-toolkit arsenal.
The real magic? Persistence. Run npx cc-agent-load standalone, it’s a terminal blip — read, discard. Via MCP, stats linger in context. Claude proactively nudges: “Streak at risk tomorrow — dial back?” Or tailors advice: “8h AI time today; let’s focus this session.”
Why Does Claude Need to Know Your Streaks?
Developers, listen up. You’ve got AI agents ghosting harder than your ex — running overnight, piling hours. Without visibility, you’re blind to the burn. cc-mcp flips that. Ask for day-by-day you-vs-AI splits. Project breakdowns reveal your black holes (looking at you, that side-project monorepo). Projections? Pure math on your pace.
Cynical aside — Anthropic’s PR would spin this as “empowering creators.” Please. It’s a local hack exposing their own black box. No corporate overlords profiting here; just a dev tired of copy-pasting stats. Run cc-health-check for a 20-point diagnostic — score under 80? Ops Kit fixes it one-shot.
Full loop: track, understand, predict, act, ask. cc-mcp welds it shut.
Is cc-mcp Worth the Hype for AI Devs?
Short answer: yes, if you’re deep in Claude Code. Skeptical long answer — it’s not revolutionary; it’s the missing pipe. We’ve had stats tools forever. The sin? They didn’t talk back. This does. Expect copycats for Cursor, Copilot. But local-only? Smart move in a world of data vampires.
Downsides? Node.js dep, config fiddling. Not for casuals. And streaks? Gamification bait — who’s really winning, you or the habit loop?
Still, in a sea of buzzword salads, cc-mcp delivers. GitHub: yurukusa/cc-mcp. npm: @yurukusa/cc-mcp. Try it. Question your habits.
🧬 Related Insights
- Read more: Rugged Edge Devices: Saving Industrial Software from Harsh Reality
- Read more: Kubernetes Debugging: Secure, Swift, and Actually Auditable
Frequently Asked Questions
What is cc-mcp and how does it work with Claude?
cc-mcp is a local MCP server giving Claude access to your Claude Code stats like hours, streaks, Ghost Days via simple config and npx.
How do I install Claude MCP server for usage stats?
npm install -g cc-agent-load, add to claude_desktop_config.json, npx @yurukusa/cc-mcp, restart app. All local, no cloud.
What are Ghost Days in Claude Code?
Ghost Days: when AI sub-agents run autonomously without your interactive sessions — perfect for tracking unsupervised AI work.
