The MCP whiteboard: let your coding agent read the plan you drew

Planloom ships an MCP server. Connect any MCP-compatible coding agent with an access token and it reads your canvas as an ordered plan: steps, arrows, frames and linked tasks. It asks questions as pins on the exact shape, proposes edits you accept one by one, and shows what it is working on while it runs.

What you get

Draw once, stop re-prompting

Arrows become order and dependencies. The agent gets the plan, the neighbours of each step and the rules of its frame, so you stop pasting context into chat.

Questions on the board

When the agent hits a fork it pins a question with options to the shape instead of interrupting you. Answer when you are ready; it picks the answer up.

Suggestions, not silent edits

Agent changes arrive as ghost previews in a tray. You accept or reject each one, and accepting is a single undo step.

Live agent presence

Outlined shapes and a status chip show which agent is on which step, and when one is waiting on you or blocked.

Scoped access

Tokens are created in your settings, and read-only tokens can read the plan without being able to change anything.

How it works

  1. 1Create an access token in Settings, under AI access.
  2. 2Add the Planloom MCP server to your agent with that token.
  3. 3Point the agent at a canvas and watch it read the plan, ask, and report progress.

Questions

What is an MCP server?

MCP (Model Context Protocol) is an open standard that lets AI agents use tools and data from other apps. Planloom exposes your canvas and tasks through it.

Can the agent edit my canvas directly?

It files suggestions that you review. Nothing is applied until you accept it, and each acceptance is one undo step.

Does it work on any canvas?

Reading plans works on Co-Space canvases, which is also where questions, suggestions and live presence appear.

Plan software visually. Ship it from GitHub.

Open a canvas in seconds. No account, no credit card.