CCC — Coding Agent Command Center
One local board that shows which of your nine coding agents needs you right now, and lets you answer without opening the rest.
- Source repo
- amirfish1/claude-command-center
- Stars
- ★ 217
- Last updated
- today
- License
- NOASSERTION
- Primary language
- Python
- FA score
- 50/100 · Major gaps
At a glance
- Works with
- Portable with changesCodex · Claude CodeOpenAI API (Partial support)
- You'll need
- Typical use
- Developers juggling several Claude Code and Codex terminals who keep losing track of which session is blocked on a question.
- Main limitation
- Licensed under FSL-1.1-MIT rather than plain open source: you may not sell CCC or offer it as a competing or hosted service, and each release converts to MIT only two years after shipping.
- Source review
- 50/100 · Major gaps
What does this agent do, and when should you use it?
CCC is a local dashboard and coordination layer for coding-agent sessions. Instead of wrapping or migrating anything, it reads the session files that Claude Code, Codex, Cursor, Antigravity, Kilo Code, Kimi Code, OpenCode, Grok and Devin already write, so sessions started from a terminal, an IDE or CCC itself land on a single board at http://localhost:8090. Each row reports whether an agent is working, waiting or finished, and when a session stops on a question you get the options with short summaries and can answer in one tap. Beyond the list it ships a Flow canvas and project tree for grouping, split-pane conversation reading, cross-session search, group chat for agent-to-agent coordination, and durable WatchTower-backed queues with workers that accumulate learnings. It runs on Python 3.9+ and Git, stays local by default, and offers redacted share cards and run replays generated on your machine.
CCC scans on-disk session data produced by each agent (for example Claude Code session files) and renders it as a unified session list plus a conversation reader; sessions started with a CLI command such as claude are picked up automatically. When an agent halts for input, CCC parses the question and presents the options with a per-option summary, then writes your choice back so the session continues. The Flow canvas and project tree group live sessions by the work they belong to, and dragging a session onto the edge of an open conversation opens it side by side with its own input bar. Search spans session history, with an optional semantic mode. Group chat delivers one message to every participant so answers no longer travel between terminals. Queues are powered by the bundled WatchTower engine: tickets survive closed sessions and parallel workers claim them, reading the queue's shared learnings before a task and writing back after. Sharing is local: Share this run or ccc share <session> produces a redacted replay page or image, and the fleet share card reports totals only — no prompts, code or project names.
- Developers juggling several Claude Code and Codex terminals who keep losing track of which session is blocked on a question.
- Teams that want an Opus-class planner to slice work for Codex, Gemini (via Antigravity), Grok or Devin while reviewing every diff in one place.
- Solo builders who hit a Claude usage limit mid-fleet and want remaining sessions to continue on a free model through their own router.
- Engineering groups batching work into durable WatchTower queues so workers can pick up tickets in parallel even after sessions close.
- Remote workers who want to monitor and reply to sessions from a phone over Tailscale.
- Anyone who needs to show a colleague what a run did without exposing prompts, code or project names.
How do you install or deploy this agent?
Git and Python 3.9+ are required; macOS and Linux are the primary targets, and native Windows has its own path below. The one-line installer opens http://localhost:8090 and asks before installing a background service:
curl -fsSL https://raw.githubusercontent.com/amirfish1/claude-command-center/main/scripts/install.sh | CCC_FROM=readme bashOn macOS you can also use Homebrew, or download the DMG from the releases page:
brew install amirfish1/tap/ccc && cccOn Windows, use PowerShell:
irm https://raw.githubusercontent.com/amirfish1/claude-command-center/main/scripts/install.ps1 | iexFrom source:
git clone https://raw.githubusercontent.com/amirfish1/claude-command-center/main/scripts/install.sh
cd claude-command-center
./run.shOn Linux, register a systemd user service for a persistent install:
./run.sh --install-serviceThe installer sends one anonymous daily ping; disable it with:
CCC_TELEMETRY_DISABLED=1 ./run.shHow do you use this agent?
After install, open http://localhost:8090 and pick a repository. Existing Claude Code history appears on the board immediately, and anything you launch with claude shows up by itself. The first-run wizard can check the machine, help install Node and the Claude Code CLI plus a local free-model router, then run a first real task on a $0 model. To preview without installing, the official live demo runs in a browser on fake data. Ask for a local reminder about what your sessions did tomorrow morning with:
ccc remindShare a run from the session menu or via CLI:
ccc share <session>Phone access is configured through Settings > Phone access, which walks you through Tailscale. CCC asks before adding question-numbering guidance, hooks or skills to your global agent config, and Settings lets you review the diff or remove them later.
What are this agent's strengths and limitations?
- Attaches to work you already have: it reads each agent's own session files across nine engines, so nothing is wrapped or migrated and agents keep running after the dashboard closes.
- Turns blocked sessions into a glanceable state: questions are surfaced with option summaries and answered in one tap.
- Concrete free path: the wizard installs Node, the Claude Code CLI and a local free-model router, then reports what the run would have cost at API prices.
- Coordination features beyond a list — Flow canvas, split conversations, group chat, and durable WatchTower queues where workers read shared learnings before a task and write back after.
- Sharing is generated locally, with redacted replay pages and totals-only stats cards rather than uploaded transcripts.
- Documented paths for macOS, Linux, native Windows, WSL2, Docker and a systemd user service, with a browser demo for evaluation first.
- Licensed under FSL-1.1-MIT rather than plain open source: you may not sell CCC or offer it as a competing or hosted service, and each release converts to MIT only two years after shipping.
- Anonymous daily telemetry is on by default and must be disabled explicitly with CCC_TELEMETRY_DISABLED=1.
- Engine capabilities are uneven: Kilo Code has no resume, Cursor IDE sync is metadata-only, and GitHub Copilot CLI, VS Code Copilot Chat and Devin cloud sessions are read-only.
- Install needs Git, Python 3.9+ and at least one agent CLI; the free-model path leans on third-party free providers with their own limits, and the keyless one logs prompts for training.
- Queues and several coordination features depend on the separate WatchTower project rather than being self-contained.
- Native Windows runs in the foreground only; background service mode requires WSL2.
How does this agent compare with similar options?
The README positions CCC as a session dashboard and coordination tool and notes that model routers are the backends it consumes, not replacements for it. It does not name specific competing dashboards, so no head-to-head comparison is made; the live demo lets you judge the UI before installing.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| CCC — Coding Agent Command Center This agent | 50 · Major gaps | — | ★ 217 | today | Python | Codex · Claude Code |
| Delegatus Agent Workbench | 79 · Good | CLI | ★ 37 | today | TypeScript | Codex · Claude Code |
| Chorus Agent Harness | 72 · Some gaps | CLIFree + model costs | ★ 1.2k | 3d ago | TypeScript | Codex · Claude Code |
| Claudexor | 76 · Good | CLIFree + model costs | ★ 523 | today | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
SECURITY.md documents loopback-only default bind, same-origin POST checks, optional PIN gate, per-item consent for agent-config writes with backups and revoke, and the cloud-relay doc specifies a closed capability list with local fail-closed validation — genuine positives. However the same SECURITY.md admits headless sessions run with --dangerously-skip-permissions, /api/open no longer restricts paths, and the ACP adapter accepts arbitrary working directories, which materially weakens least privilege. Sensitive state relies on umask with a documentation-only chmod suggestion. Dependencies are claimed stdlib-only, but WatchTower is a hard dependency installed via shallow clone/tarball/PyPI fallback without pinned verification. External effects include shelling out to many CLIs and the card-worker upload path. Rollback is comparatively strong (config-backups, consent revoke). Attribution exists via LICENSE/NOTICE and author name, but publisher identity is unverified, so it is not scored high.
Version numbers disagree across files (pyproject 5.43.0, install.md wheel 5.37.0, README screenshot v5.35), indicating manual sync; WatchTower is a hard dependency with multi-stage fallback, so availability depends on external repos and network; failure messaging is comparatively specific in install.md, configuration-history.md and usage_db/cli.py with explicit exit codes and error strings.
README and install.md cover macOS, Linux, Windows, WSL2, Docker and multiple installers with clear scenarios; capability boundaries are stated (Kilo launch-only, Cursor metadata-only, Copilot read-only, etc.); however skill trigger descriptions are broad natural-language phrases and several skills assume a locally running CCC with the Bash sandbox disabled, leaving environment fit to the user.
Install documentation is thorough (curl/brew/DMG/PowerShell/uvx/pipx, service install/uninstall, log locations) and information architecture is clear; naming stability is questionable (tap renamed, version drift, stale README screenshot); examples and FAQ exist but are scattered; known limitations are honestly disclosed in SECURITY.md and install.md; the license is FSL-1.1-MIT, a non-OSI source-available license with NOASSERTION metadata, requiring readers to interpret restrictions; maintenance responsibility points to a single author and the publisher is unverified.
The product purpose (a fleet board for multiple coding-agent sessions) is clear, with dashboard, CLI, API and skill wrappers as outputs; marginal value is high for multi-session users and limited for single-session users; skills explicitly state spawned-session counts and require telling the user before spawning, but overall token cost still needs user judgment, and free-model routing carries a training-data caveat.
Some security and privacy claims map to included code (card-worker CORS/rate-limit/PNG validation, usage_db CLI), but many claims (cloud-relay encryption and retention) point to normative sources not in this repository and cannot be cross-checked; CI only compiles, smoke-tests and runs unit tests, and testing is described as minimal by design; fact and inference are mostly separated in prose, but several performance and security guarantees lack reproducible evidence.
- Headless sessions run with --dangerously-skip-permissions, so anything that can reach the API can run arbitrary commands; do not expose the port to a network.
- WatchTower is a hard dependency installed via shallow clone or PyPI fallback without pinned verification; review and pin it yourself.
- The license is FSL-1.1-MIT (non-OSI) and forbids competing commercial use; metadata is NOASSERTION, so read the terms before commercial use.
- Publisher identity is unverified and version numbers disagree across files; confirm the actual release before updating.
- Free-model routing may use prompts for training; the docs warn against sending private code.
Maintain this project? Add the FARS badge to your README
[](https://followagents.com/en/agents/claude-command-center)<a href="https://followagents.com/en/agents/claude-command-center"><img src="https://followagents.com/badge/claude-command-center.svg" alt="FARS review"></a>Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
CCC v5.43.0 The decisions waiting on you, in one place - Decisions inbox: review queue owner questions and explicitly promoted session questions together, with visible source coverage and delivery states. Answer controls are opt-in and keep Save, Send, and receipt checks separate. - A clearer session workspace: Active, Needs you, and Finished groups, collapsible categories, and explicit archive actions help you track work through completion. …View on GitHub ↗ -
Release notes excerpt
Added - Ticket rows show an AUTO switch under the avatar: purple and moving while CCC drives the work, amber when it is waiting on you, faint when you drive it. Click it to hand off or take back. - Child lanes show only a tiny model icon instead of a repeat of the parent's numbered avatar. - Features and Small bugs now manage themselves (idea tickets). A finished session closes its ticket with its own summary and moves to Archived. A session that still has work it owns is handed off: CCC tells it to drive the ticket to a verified close and watches it, nudging a stalled session and turning …View on GitHub ↗ -
Release notes excerpt
Changed - The macOS app can keep the CCC server running in a background service (asked once on first launch). Quitting, relaunching or updating the app no longer stops the server or live streams, launchd restarts it after a crash, and the CCC menu has Restart Server and Stop Server and Quit. The dashboard launch agent is now com.github.claude-command-center.server, so it no longer collides with the app's bundle ID; an older install is migrated automatically and its plist is kept in ~/.claude/command-center/service-backup/. New ./run.sh --start-service and --stop-service. …View on GitHub ↗
FAQ
Does it replace Claude Code or Codex?
Is it really free?
Where does my data go?
Which agents are only partially supported?
Can I use it from my phone?
Thanks — we use this to improve the page.