Cielo
A Kanban board built for AI agent orchestration. Agents claim tasks, track dependencies, and coordinate work through a built-in MCP server — while humans watch it all happen in real time.
Overview
Cielo is a Kanban board designed for multi-agent AI workflows. It looks like Trello — boards, lists, cards, drag-and-drop — but every feature is built with AI agent interaction as a first-class concern. A built-in MCP server exposes 20 tools that let agents read board state, claim tasks, manage dependencies, and update progress programmatically.
Features
- MCP server — 20 JSON-RPC tools for full board manipulation by AI agents
- Actor attribution — every mutation is logged with who did it, enabling multi-agent debugging
- Card dependencies — directed graph of blocking relationships between tasks
- Real-time sync — Server-Sent Events push updates to all connected clients instantly
- Drag-and-drop UI — familiar Kanban interface for human operators watching agent workflows
- Priority and status tracking — cards flow through unassigned, assigned, in_progress, blocked, and done
- Full-text search — filter by title, assignee, status, or label within a board
- Single binary — Go backend, React frontend, and SQLite database in one deployable artifact
Architecture
Cielo runs as a single Go process serving the REST API, MCP server, and embedded React frontend. SQLite with WAL mode handles persistence — no external database needed. An internal event bus decouples mutations from real-time broadcast via SSE.
The MCP server shares the same service layer as the REST API, so agents and the web UI have identical capabilities. Agents identify themselves via an actor parameter on every write operation, creating an immutable audit trail.
Getting Started
# Build and run
make build && make run
# Or with Docker
docker compose up -d
The board is available at http://localhost:8080. Point your MCP client at the same address to connect agents.