ccli

A unified CLI for inspecting your Claude Code installation — MCP servers, skills, rules, projects, and metadata.

Why ccli?

Claude Code stores its configuration across many files and directories — global settings, project configs, MCP server definitions, skills, rules, session history, and more. There's no built-in way to get a unified view of what's configured, where it lives, or how it all fits together.

ccli gives you that visibility in a single command-line tool:

• See everything at a glance — ccli info shows your full setup: version, auth, model, paths, session counts, and storage usage.
• Audit MCP servers — List servers across all scopes, inspect their config, and verify environment variables without digging through JSON files.
• Discover skills and rules — Find what's available across personal, project, and plugin sources in one place.
• Track project usage — View per-project costs, token usage, and model breakdowns from session history.
• Clean up old sessions — Delete old session data and associated artifacts (debug logs, telemetry, todos, tasks) across all or specific projects.
• Scriptable output — Every command supports --format json and --format yaml for automation and piping.

All of this works offline by reading config files directly — no network calls, no dependency on the claude binary (except for version/auth detection).

Looking ahead, ccli is designed to grow beyond Claude Code. As the ecosystem of AI coding agents expands — Cursor, Windsurf, Codex, and others — each brings its own configuration formats, MCP setups, and project conventions. ccli aims to become a single pane of glass for inspecting and managing configuration across multiple agents, so you can understand your full AI-assisted development setup regardless of which tools you use.

Installation

Homebrew

brew install jackchuka/tap/ccli

Go

go install github.com/jackchuka/ccli@latest

Usage

Info dashboard

ccli info

Shows version, auth status, model, paths, session/project counts, storage size, and resource counts.

MCP servers

# List all MCP servers across all scopes
ccli mcp list

# Show details for a specific server
ccli mcp get datadog

Displays server type, command, URL, and environment variables (sensitive values are masked).

Skills

# List all skills grouped by scope
ccli skills list

# Show details for a specific skill
ccli skills get brainstorming

Discovers skills from personal (~/.claude/skills/), project (.claude/skills/), and plugin (~/.claude/plugins/cache/) sources.

Rules

# List all rules from global and project scopes
ccli rules list

# Show details for a specific rule
ccli rules get code-comments

Projects

# List all known projects with usage stats
ccli projects list

# Show detailed stats for a project
ccli projects get my-project

Displays per-project cost, token usage (input/output), line changes, session count, and per-model cost breakdown.

Cleaning up sessions

# Delete sessions older than 30 days across all projects
ccli projects clean --older-than 30d

# Preview what would be deleted
ccli projects clean --older-than 30d --dry-run

# Clean a specific project
ccli projects clean my-project --older-than 7d

Removes old session logs and associated artifacts (debug logs, telemetry, todos, tasks, file history, session environment) matched by session UUID.

Output formats

All commands support --format for machine-readable output:

ccli mcp list --format json
ccli skills list --format yaml
ccli info --format json

Flags

How it works

ccli reads Claude Code configuration files directly from disk:

All resources are categorized by scope — global, project, personal, or plugin — shown with colored bullets in text output.

No network calls. No dependency on the claude binary (except for version and auth status).

License

MIT