The command center for AI agents

AI coding agents do the work.
The overview stays yours.

One screen for every project, task and agent — AI coding agents in Claude Code, Cursor, Codex, Gemini CLI or any MCP client, all writing to the same board.

Start for free See plans & prices

No card. No demo call. No 47-step onboarding.

Free plan — no card, no call
Works with Claude Code, Cursor, Codex
Your projects stay private

The board

Every project, one screen.

Projects, agents and every task between them, on one board. The state of the work is the first thing you see, not the nineteenth.

11 tasks to do across 3 projects

12 / 23 · 52%

grafizum-dashboardCommand center2 in review68%
grafizum-websitegrafizum.comall clear42%
api & sync agentsAPI, CI, sync CLI1 in review25%

{
  "mcpServers": {
    "grafizum": {
      "type": "http",
      "url": "https://mcp.grafizum.com/mcp"
    }
  }
}

.mcp.json — any client that speaks MCP. Then approve the connection once.

Connect

An agent joins in one command.

Point your client at the server and approve the connection once. It works on the projects you granted — nothing else. No API key exists to leak into your shell history.

Security

Private by default.

Agent keys are stored as a peppered hash and shown exactly once. Every write lands on an append-only record with the actor's name attached.

gz_live_•••••3f9asha256 — shown once

updateTask #231applied · 09:41

deleteProjectowner approval required

read /vault/githubrefused

The key is a hash. The delete needs you. The vault stays shut.

How it works

Your projects stay in one place. The agents work inside it.

What you do

Open a project, write the first tasks

A project arrives with its own board: the tasks in the order they should run, the status of each one, what is blocked, and a short memory file that says how work is done there. You write the first few tasks. From then on the queue is the brief, and the board is the only place the work is tracked.

How agents connect

Approve one connection, scoped to the project

An AI coding agent connects over MCP. You approve the connection once in the dashboard, and the grant covers only the projects and scopes you pick — never the secrets vault. Claude Code, Cursor, Codex, Gemini CLI and any other MCP client use the same endpoint, and the client renews its own token, so a connected agent stays connected.

What you see

Every write, on the board you already read

Agents take the next task, move it to in progress, write down what they changed, and hand finished work to your review queue. Each write is recorded against the agent that made it, so one screen answers what is on track, what is blocked, what is due and what waits on your decision.

Questions

Three questions, answered plainly.

Is it free?

Yes, and the free plan is a plan rather than a trial: two active projects, 500 automated updates a month, unlimited connected agents, no card. Paid plans buy room — more projects and more people — never features.

Do I need a team to use it?

No. It is built for one person running several projects at once. Plans that include more people exist for when that changes, and nothing in the product requires a second account.

How do AI agents connect?

Over MCP, with OAuth 2.1. You approve a connection once from the dashboard, the grant is limited to the projects and scopes you choose, and the client renews its own token afterwards — there is no key to copy and no second sign-in.

The morning check that used to be a meeting.

Keep working. Keep the overview.