See everything Claude Code knows — projects, MCP servers, sessions, costs — in one dashboard.
Setup Guide | Security | Contributing | Roadmap | Changelog
A local dashboard for visualizing and managing your Claude Code ecosystem. Auto-discovers your projects, MCP servers, skills, plugins, sessions, agents, and their relationships with zero configuration.
- "How much am I spending on Claude Code?" -- Cost analytics by session, project, model, and day. Set a monthly budget and get 80%/100% alerts. Opus→Sonnet savings recommendations.
- "Which MCP servers are configured where?" -- Auto-discovers every
.mcp.jsonacross all projects. One view, zero setup. Stack-aware recommendations for what's missing. - "I had a session last week that fixed this exact bug..." -- Deep search across all session content. Find any conversation by what was said, not just the title.
- "What was I working on yesterday?" -- Continuation intelligence detects unfinished work and generates context prompts to resume instantly.
- "I have 300+ sessions eating disk space." -- Stale detection, bulk delete, and health scores help you clean up. Pinned sessions survive bulk delete.
- "My graph has 1000+ edges and looks like spaghetti." -- Focus Lens shows one node's neighborhood at a time in a clean radial layout, with a searchable picker for any entity in the full graph.
| Dashboard -- entity counts, monthly budget, insights, system health, quick actions | Sessions -- deep search, AI summaries, pins, notes, HTML export, delegation |
| Live View -- real-time sessions with context bars, compact button, cost ticker | Analytics -- daily cost charts, model breakdown, CSV export, budget tracking |
| Graph -- Focus Lens, 6 view modes, blast radius, custom nodes, AI suggestions | MCP Servers -- auto-discovered, categorized, with stack-based recommendations |
| Help Center -- 107 topics, glossary, cheat sheet, first-5-minutes walkthrough | Prompts -- save and reuse effective prompt templates with tags |
# Clone and run (3 commands)
git clone https://github.com/sorlen008/claude-command-center.git
cd claude-command-center
npm install && npm run devOpen http://localhost:5100. Done. Everything is auto-discovered from your ~/.claude/ directory — no configuration needed.
Tip: Keep it running in the background while you work in Claude Code. The dashboard auto-refreshes every 3 seconds.
See SETUP.md for npm global install, production builds, and troubleshooting.
- Node.js 18+ (tested on 20, 22, 24)
- Claude Code installed -- the dashboard reads from
~/.claude/which Claude Code creates - git -- required for the update feature (optional otherwise)
- Auto-discovers all Claude Code projects, MCP servers, skills, plugins, and markdown files
- Help Center -- four-tab learning surface with 107 topics, 49-term glossary, cheat sheet, first-5-minutes walkthrough, per-topic beginner/advanced difficulty filter, URL-hash deep-linking
- Session intelligence -- deep search, AI summaries, cost analytics, file heatmap, session health scores, HTML export
- Automated insights -- cost-spike detection, Opus→Sonnet savings suggestions, stale session alerts, duplicate-work detection, budget warnings
- Monthly budget planner -- set a spend cap, watch it on the Dashboard, get 80% and 100% alerts
- Graph Focus Lens -- survival mode for dense graphs (200+ nodes): shows a focus node + its neighborhood in a clean radial layout, searchable focus picker, escape hatch to show all
- Graph visualization -- 6 view modes (force / tiles / tree / list / radial / matrix), blast radius, custom nodes via
graph-config.yaml, AI-assisted suggestions - MCP recommendations -- analyzes project tech stack and suggests matching MCPs from the catalog
- Operations nerve center -- real-time service health, cost pacing, attention items, overnight activity
- Continuation intelligence -- detects unfinished work, uncommitted changes, abandoned sessions
- Bash knowledge base -- every shell command indexed and searchable with success rates and failure hotspots
- Decision log -- AI-extracted architectural decisions with alternatives and trade-offs
- Natural language query -- ask questions about your analytics data ("Which project costs the most?")
- Session delegation -- continue sessions via terminal, Telegram, or voice (cross-platform)
- Project dashboards -- per-project cost, health, files, and session aggregation
- Prompt library -- save and reuse effective prompt templates
- Agent tracker -- definitions and execution history across sessions
- Live view -- real-time monitoring with context bars, compact button, click-to-open in Sessions, cost ticker
- Markdown editor -- edit
CLAUDE.mdand memory files with autosave, version history, drift analyzer, overlap detection - Discovery -- finds unconfigured projects and MCP servers on disk
- One-click updates -- check and apply updates from the sidebar
- Mobile-responsive -- sidebar collapses below 768px, accessibility labels throughout
| Page | Description |
|---|---|
| Dashboard | Entity counts, monthly budget, automated insights, usage card, quick stats |
| Projects | Discovered projects with session counts, tech stack, per-project dashboards |
| MCP Servers | Every MCP server found in .mcp.json files, categorized, with stack-aware recommendations |
| Skills | Auto-categorized by inferred keywords (devops, quality, docs, ai, etc.), favoriting, grouping |
| Plugins | Installed plugins with their bundled commands, agents, skills, and hooks |
| Markdown | CLAUDE.md, memory files, READMEs — autosave editor, version history, drift analyzer, overlap detection |
| Sessions | Deep search (three modes), AI summaries, pins, notes, HTML export, bulk delete with undo, delegation |
| Agents | Agent definitions, execution history, stats, learn guide |
| Live | Real-time sessions with context bars, compact button, click-to-open in Sessions |
| Graph | Interactive node graph — Focus Lens default, 6 view modes, blast radius, custom nodes, AI suggest |
| Prompts | Reusable prompt templates with tags and favorites |
| Messages | Cross-session message timeline with content search |
| APIs | External APIs from apis-config.yaml, env files, and Docker Compose |
| Activity & Discovery | File-change timeline plus auto-detected infrastructure (Docker, git remotes, env URLs) |
| Stats | Cost analytics, daily charts, model breakdown, CSV export, weekly digest |
| Settings | Path config, app name, monthly budget, entity overrides, themes |
| Help Center | 107 topics across 15 categories, glossary, cheat sheet, first-5-minutes walkthrough |
The Sessions page includes a full Analytics tab with:
- Cost Analytics -- total spend, per-model/project/day breakdowns, most expensive sessions
- File Heatmap -- most-touched files with read/edit/write counts, clickable for cross-session timeline
- Session Health -- tool error and retry pattern detection (good/fair/poor scoring)
- Bash Knowledge Base -- searchable index of every shell command with success rates
- Decision Log -- AI-extracted architectural decisions from past sessions
- Operations Nerve Center -- configurable service monitoring, cost pacing, attention items
- Continuation Intelligence -- unfinished work detection with one-click resume
- Smart Context Loader -- generates context prompts from recent session summaries
- Natural Language Query -- ask questions about your data using Claude Haiku
- Prompt Library -- save reusable templates with one-click copy
- Weekly Digest -- automated weekly summary with accomplishments
- Auto-Workflows -- configurable auto-summarize, stale detection, cost alerts
This tool runs entirely on your local machine.
| Concern | Details |
|---|---|
| File system | Reads ~/.claude/ and project directories. Writes only to ~/.claude-command-center/ and markdown files you explicitly edit. |
| Shell commands | Spawns claude -p, git, platform file openers, terminal emulators. All user input validated with Zod. |
| Network | Binds to 127.0.0.1 only. No outbound requests unless you use Discovery search or AI Suggest. |
| Data | All data stored locally as plain JSON. No cloud sync, no external databases. |
| Telemetry | None. No analytics, no tracking, no phone-home. |
| Secrets | Never stored. Scanned env vars with "secret", "password", "token", "key" are redacted to ***. |
See docs/security-threat-model.md for the full threat model.
| Variable | Default | Description |
|---|---|---|
PORT |
5100 |
Server port |
HOST |
127.0.0.1 |
Bind address. Do not set to 0.0.0.0 -- no authentication. |
COMMAND_CENTER_DATA |
~/.claude-command-center/ |
Data directory |
GITHUB_TOKEN |
(none) | Optional. GitHub API rate limits for Discovery. |
NERVE_CENTER_SERVICES |
Command Center:5100 |
Services to monitor (name:port,name:port). |
VOICE_CALLER_SCRIPT |
(disabled) | Path to voice outbound caller script for delegation. |
VOICE_PHONE |
(disabled) | Phone number for voice delegation. |
Extend the auto-discovered graph with custom nodes via graph-config.yaml:
nodes:
- id: my-database
type: database
label: "PostgreSQL"
description: "Primary database on :5432"
edges:
- source: my-mcp-server
target: config-my-database
label: connects_toPlace in ~/, ~/.claude/, or any project directory. See SETUP.md for details.
AI Suggest -- click the button in the graph toolbar to get AI-generated suggestions for infrastructure nodes and connections. Requires Claude Code CLI.
npm run build # Bundle client (Vite) + server (esbuild)
npm start # Run production bundleThe sidebar shows update indicators. Or manually: git pull && npm install && npm run build.
curl -LO https://github.com/sorlen008/claude-command-center/releases/download/vX.Y.Z/claude-command-center-vX.Y.Z.tar.gz
curl -LO https://github.com/sorlen008/claude-command-center/releases/download/vX.Y.Z/checksums-vX.Y.Z.sha256
sha256sum -c checksums-vX.Y.Z.sha256Frontend: React 18, TanStack Query, Tailwind CSS, Radix UI, React Flow Backend: Express 5, chokidar, Zod Build: Vite + esbuild, TypeScript throughout No external services -- everything runs locally
See CONTRIBUTING.md. Security issues via SECURITY.md.