Turn solo AI agents into a coordinated engineering team
Intelligent workflow orchestration for AI coding agents. Transforms Cursor, Copilot and Roocode into structured dev teams with persistent context, task tracking, and role-based guidance.
AI coding assistants produce inconsistent output because each prompt starts from scratch — no shared context, no role boundaries, no quality gates. Anubis installs a persistent workflow layer that gives Cursor, Copilot, and Roocode structured roles, task memory, and review checkpoints — so they work like a team instead of a single overloaded tool.
The challenge
Giving stateless AI agents persistent memory and structured roles without modifying the agents themselves.
The outcome
Developers using Anubis report more consistent, reviewable AI output — with full task history that survives IDE restarts.
Technical approach
- MCP protocol — Model Context Protocol, the open standard Anubis exposes to connect AI clients
- Persistent context — workflow state survives across agent sessions via structured storage
- Role-based agent guidance — architect, developer, reviewer personas with distinct responsibilities
- Task lifecycle management — create, claim, progress, review, complete state machine
- npx server bootstrap — zero-install invocation via npx for Cursor / VS Code MCP config
- Structured task queue — JSON-serialised task records with priority and dependency tracking
- Multi-agent coordination — sequential and parallel handoff patterns across agent personas
- Context injection — automatic project metadata injected into every agent prompt
Results at a glance
- ★ 124
- GitHub Stars
- 7k+
- npm installs
