Workflowy MCP vs listlis: local bridge or hosted connector?

Both Workflowy and listlis let an AI client read and change an outline through the Model Context Protocol. The difference is where that connection lives. Workflowy’s Desktop MCP runs through the signed-in desktop app on your computer. listlis exposes a hosted MCP endpoint that you connect to with your account.

Workflowy Desktop MCP fits a person who already works in the desktop app and wants a local AI tool to work on that same machine. listlis fits a person who wants an AI client to connect through a browser sign-in, without a desktop app staying open as the bridge. Both paths can search, read, and write an outline. The better question is what you need the connection to carry.

Workflowy Desktop MCP keeps the bridge on your computer

Workflowy’s Desktop MCP is built into its desktop app. Install or update the app, sign in, enable Local MCP in Settings, then copy its generated Connection Link into an MCP client that supports Streamable HTTP. The app authenticates with the Workflowy account already signed in there, so this path does not ask for a separate API key.

That has a clean operational boundary. The connection works while the desktop app is open and Local MCP is enabled. Close the app or disable Local MCP and the client loses access. Workflowy recommends treating the Connection Link like a private key, which is the right instinct for any URL that gives an AI access to your notes.

For a lot of work, that is all you need. A local client can search an account, read a project, add the action items that came out of a chat, move finished work into an archive, or reorganize a messy branch. The connection is close to the place where the outline already lives.

Workflowy also has a separate Local MCP app. It is a different setup, not merely a setting in the desktop app. It requires a Workflowy API key and maintains a local SQLite cache. Its extra features include fast full-text search, multiple accounts, backups, bookmarks, and persistent AI instructions. That option is for someone who wants to run more of the connection on their own computer and is comfortable maintaining another app.

The distinction matters because “Workflowy MCP” can mean either path. Desktop MCP is the shorter setup. The Local MCP app adds local data and a few more moving parts in return for its cache and account-management features.

listlis hosts the connection and signs the client in with OAuth

listlis puts its MCP server at https://mcp.listlis.com/mcp. A compatible client connects to that URL and opens a browser sign-in for the listlis account. The connection is an OAuth grant, so it belongs to the account rather than to a desktop app running in the background.

That changes the setup more than the protocol. There is no local server to keep running and no connection link copied out of a desktop app. The trade is that the account and the client need to complete the OAuth sign-in flow. The listlis connection guide has the same two-minute setup for Claude and Claude Code.

The hosted endpoint is useful when an assistant needs to work with the same outline from a compatible client without depending on one computer being awake. It is also a better fit when the outline is already a shared, browser-first workspace and the desktop app is not part of the daily routine.

The connection model does not make the outline less personal. The client still receives access through the account, and it should still be a client you trust with the material it reads or writes. The practical difference is where you manage that access: a local bridge in Workflowy, or the account connection in listlis.

The tree context is where listlis gets specific

An AI can do useful work with a simple read-and-write connection. It gets more useful when it can see where a line belongs.

listlis starts an AI at the top of the tree, then lets it ask for a focused node with the Markdown body, the ancestor path above it, its siblings, and the branch below it. Search results include their breadcrumb. That means an assistant does not have to treat a matched line as an orphaned note.

Imagine a project that looks like this:

Fall launch
  Website
    Rewrite pricing page
      Notes from Tuesday's call
    Collect screenshots
  Follow-ups
    Ask Jordan about the case study

If you ask for the pricing work, a connected client can see that it belongs under Website, that screenshots are a nearby concern, and that the meeting notes sit below the task. It can add a child note, create a next action in the right branch, or move a section without turning the project into a flat list of search hits.

listlis also exposes a due-date agenda with overdue, today, and next-seven-day sections. Its topic-pack tool searches the tree and returns the relevant branches as a single Markdown bundle, with breadcrumbs and a token budget. For a research question or a project handoff, that is often a more useful first step than pasting one search result into a chat.

On the write side, a client can create one node, import a whole Markdown outline as a subtree, update a node, and reparent or reorder a branch. The server calculates sibling placement, so an AI asks for “after this item” or “last child” instead of inventing ordering data. The public MCP page walks through the same tools from the person using the outline.

What the connection choice changes

Question	Workflowy Desktop MCP	listlis hosted MCP
Where does the bridge run?	In the signed-in Workflowy desktop app on your computer	At mcp.listlis.com, connected through account OAuth
What must stay available?	The desktop app and Local MCP setting	The listlis account and the connected MCP client
How does setup start?	Enable Local MCP and copy the generated Connection Link	Add the server URL and sign in in the browser
What does it suit?	Working with an existing Workflowy account from a local AI client	Working with a browser-first listlis tree through a hosted endpoint
What is the context model?	Workflowy documents read, search, navigate, and update access	Nodes can include their path, siblings, subtree, agenda context, and packed topic branches

This is not a feature-count contest. Workflowy’s connection is a good route when Workflowy is already the place where you work and the desktop app is part of that routine. listlis is built around a different working surface: one keyboard-first tree for notes, tasks, and projects, with an MCP tool set that preserves the tree around a request.

Pick the connection after you pick the outliner

MCP should support the way you already organize work. It should not become the reason to move an outline that is otherwise doing its job.

Workflowy Desktop MCP is a direct fit if the Workflowy desktop app is already open while you work and you want a local client to help search, summarize, capture, and reorganize that account. Its Local MCP app adds a separate option for people who want a cache, backups, or multiple accounts on the machine.

listlis is for the person who wants the outline itself to stay simple: a title, an optional Markdown body, and as much nesting as the work needs. The hosted connection makes that same tree available to an AI client, including the branch around a node and a due-date view of open work. It does not turn listlis into an AI workspace. The outline remains the working surface.

If you are weighing the wider products, the outliner comparison hub maps the honest cases for Workflowy, Logseq, Dynalist, Tana, and Roam. The Workflowy alternatives guide covers the broader decision, and the 100-node limit guide is for the separate question of whether Workflowy’s Basic plan still fits your capture habit.

Open a root in listlis and connect an AI client to one live project. A few real tasks and notes will show whether the hosted tree is the shape you wanted.