firstintent/ccteam

What it solves

Many coding AI agents (like Claude Code, Codex, Grok, and Kimi) operate in isolation, meaning they lack a shared context, a unified identity, and a way to collaborate. ccteam provides the "connective tissue" to turn these separate CLI tools into a coordinated team, allowing users to manage multiple agents from a single console and enable agents to delegate work to one another.

How it works

ccteam acts as a bridge and daemon that orchestrates various coding CLIs. It provides a unified interface—via a web console, Telegram, or Lark—to spawn, dispatch, and collect work from different AI vendors on different machines.

Key mechanisms include:

• Multi-Vendor Orchestration: It manages sessions across different providers (e.g., Claude, Codex, Grok, Kimi, DeepSeek Harness) and allows them to be routed based on their strengths.
• MCP Tools: It provides six Model Context Protocol (MCP) tools that allow agents to spawn, read, and stop other agents, and check the status of the team.
• Cross-Machine Execution: Users can register "satellites" to run agents on different machines (e.g., a GPU box) while managing them from one central console.
• Flows: For structured work, users can write JS scripts (flows) to define pipelines, parallel tasks, and phases of execution.
• Policy Hooks: Executable scripts that gate every delegation request to ensure guardrails and budget control.

Who it’s for

Developers who use multiple AI coding assistants and want to move beyond single-session interactions to a multi-agent system where different models can be used for planning, implementation, and review in a coordinated fashion.

Highlights

• Unified Control Plane: Manage all agents via a web UI, Telegram, or Lark, including HITL (Human-in-the-Loop) approval buttons.
• Agent-to-Agent Delegation: Agents can hire other agents in plain language to handle specific tasks (e.g., "Spawn a codex session to implement RFC-12").
• Live Topology View: A "Team page" that visualizes the delegation tree, showing who delegated whom, the model used, and the running cost ledger.
• DeepSeek Harness Integration: Native plugin for DeepSeek Harness Web, providing a workbench and engine supervisor.
• Cost and Budget Management: Includes a cost ledger and daily budget caps to prevent runaway expenses.