The missing network layer for coding agents

What it does

ccteam is a Rust daemon that registers the coding CLIs already on your machines — Claude Code, Codex, Grok, Kimi, DeepSeek Harness, Pi — and knits them into a single fleet with identity, routing, delivery guarantees, and a running cost ledger. You steer the whole swarm from Telegram, Lark, or a browser tab, dispatching work across vendors and hosts as if they were colleagues rather than strangers. Six MCP tools expose the fleet to any enrolled session or external agent that presents a credential.

The interesting bit

Instead of treating orchestration as prompt engineering, ccteam treats it as infrastructure: an executable policy hook gates every delegation with live quota data, and journaled JS flows using agent(), parallel(), and pipeline() survive daemon restarts so workers outlive the runner. A satellite host behind NAT can join with a token, keeping projects bound to the machines where the code lives while centralizing transcripts and billing.

Key highlights

• Cross-vendor, cross-machine dispatch from Telegram, Lark, or a web console, with a live delegation topology and cost ledger.
• Six MCP tools available to every registered session and any external agent that presents an enrollment credential.
• Executable policy hooks (pre-agent) scriptably gate every delegation; exit codes allow, deny, or refuse distinguishably if the script breaks.
• JS flow scripts for structured orchestration with budget caps, parallel limits, and journaled runs that resume after daemon restarts.
• Native DeepSeek Harness plugin (@ccteam/ccteam-ui) embedding a full workbench, transport layer, and engine supervisor inside DSH Web.

Caveats

• Satellite execution currently supports only Claude Code sessions; other vendors still run on the daemon’s host.

Verdict

Worth a look if you already pay for several coding agents and want one console that routes work, enforces budgets, and keeps a ledger. Skip it if you only use one CLI or prefer your agents solitary and ungoverned.