Skip to content
ThinkClear

About ThinkClear

ThinkClear is a mindmap canvas with an assistant that can build it with you, and an MCP server so the agent you already use can edit the same maps. It is open source, free to use, and free to run yourself.

What it is

A mindmap here is a tree of topics with one root, drawn on a canvas you can drag. You type into it, or you say what you are working through and the assistant grafts the branches on: creating topics, renaming them, moving them between parents, deleting the ones that turned out to be wrong. Every topic can carry one markdown note, written as source and read as rendered prose, so the thinking that does not fit in four words on a node has somewhere to live on the node it belongs to.

The part that is unusual is the second transport. The tools the built-in assistant calls are also served over the Model Context Protocol, so a client like Claude Code or Cursor gets the same twelve tools against the same maps, under the same ownership rules. An edit made from a terminal shows up on the open canvas as it happens, because every server-side write announces itself over a server-sent event stream and the editor reseeds from it.

Who publishes it

ThinkClear is an independent open-source project, not a company. There is no sales team, no seat count, and no contract to sign — which is also why there is no address on this page: there is no registered entity behind it to give one for. The whole source tree is public, including the deployment configuration and the design system, and the issue tracker is where the roadmap actually lives.

  • Source on GitHubthe monorepo — API, web app, landing page, shared schemas
  • READMEwhat it is and how to run it locally
  • License (AGPL-3.0)network use counts as distribution, so a hosted fork publishes its source too
  • Deployment guidethe two-host setup this instance runs on

How it is built

The API is NestJS on MongoDB with Better Auth, which doubles as the OAuth 2.1 authorization server the MCP endpoint authenticates against. The app is React on Vite with React Flow for the canvas. This page is Next.js, deployed separately, and it imports nothing from the rest of the repository on purpose — a copy change here should never be a reason to rebuild the API.

Everything the assistant can do goes through the same service the HTTP routes do, so there is exactly one place ownership is checked. Another person's mindmap and a mindmap that does not exist are both a 404, from every transport.

What it costs

Nothing. Sign up with an email address and a password, or with Google, and you are in — no credit card, no trial clock, no call with anyone. Connecting an agent is self-serve too: there is no key to request, because the OAuth flow issues the token itself the first time your client calls the endpoint.

The hosted instance is run at the maintainer's expense and is offered as-is. If that is not a footing you want to build on, the alternative is not a bigger plan — it is running the same code yourself, which costs a container and a MongoDB and no license.