PugBase turns an idea, or a video/PDF/doc, into a structured plan your team and your AI agents share: overview, constraints, API contract, data model, and task checklists. Connect Claude Code or Cursor over MCP - your agent reads the plan, builds it in your repo, and ticks off each task as it goes. Your team watches progress live on a status board, then adds the next feature as its own tracked increment. Human-readable, agent-executable - the plan and the build stay in one place.

Hey Product Hunt! 👋 I'm Panudeth, the maker of PugBase.

I'd plan a feature, hand it to Claude Code, and half an hour later nobody - including me - could say what it actually built.

Then a teammate started running their own agent on the same repo, and it got worse.

PugBase moves the plan out of the agent's session into a place the whole team shares.

Describe an idea, or drop in a video or PDF, and you get a real plan: constraints, API contract, data model, task checklists.

Everyone then points their own agent at it over MCP - Claude Code, Cursor, whatever they use.

Each agent claims its own tasks and ticks them off live, so two people's agents stop stepping on each other.

The part I'm most proud of: a QA agent can reopen a task with a reason, and the agent that owns it reads that reason and fixes it.

The result doesn't vanish when a session ends. At 100% it writes the handoff docs.

Free to start.

How does your team handle it today when two people both have an AI agent in the same repo? 🙏

finally a planning tool that doesn't collapse once an agent touches it, the MCP connection to Claude Code actually worked and watched tasks tick off live

@sevdaab4k Thanks Sevda! How do you handle it today when more than one agent is working the same repo? That's the part I kept getting stuck on before I built this.