Guide

Choosing a diagramming MCP server.

The differentiator is not the vendor. It is whether a language model is being asked to do geometry.

Download for macOS
v0.1.33 · .dmg · Apple Silicon & Intel
Signed & notarized by Apple · opens without a Gatekeeper warning
sha256 698955a0187bc039f4c74f5d05a9f10fbb27376a45788a0a241d1326b73873c7
Download for Windows instead
$curl -fsSL https://lucidtrain.com/install.sh | sh

A diagramming MCP server exposes diagram tools to a coding agent through the Model Context Protocol, so Claude Code, Codex, Cursor or any other MCP client can produce diagrams without leaving the session. The decision that matters most between them is not which vendor you prefer but whether the server asks the language model for coordinates. Servers exposing drawing primitives make the model choose positions, which is the thing models are worst at and the reason primitive-level output overlaps past a handful of nodes. Servers taking a semantic graph and running a layout solver remove that failure entirely.

How to read this

There is no ranking here and no winner, because the honest answer depends on constraints this page cannot see: whether your code may leave the machine, whether anyone will maintain what you write, and whether the diagram has to be reviewed in a pull request. Each entry says what it actually does and when to pick it.

The primitive-level approach

Servers around draw.io and Excalidraw typically expose create-shape and create-connector style tools. The model decides what to place and where. This works for four or five boxes and degrades quickly: shapes overlap, edges route through cards, and the result needs manual tidying, which defeats the point of generating it. No amount of prompting fixes this, because the request is asking the model to do the thing it is least reliable at.

The solver approach

The alternative is a tool that accepts nodes, edges and grouping with no coordinates at all, and computes positions with a constraint solver such as ELK or Graphviz. The model decides structure and never decides position, so overlaps are not unlikely but inexpressible. Lucid Train works this way: `lucid-train mcp` runs it as an stdio server and the tools arrive as mcp__lucid__emit_diagram and similar.

The transport limitation to check first

Most MCP clients in this space accept stdio servers, meaning a local child process with piped stdin and stdout. Remote HTTP and SSE servers are frequently discoverable in a client's registry and not actually connectable, which produces a confusing first-use failure. Lucid Train's MCP support is stdio only and says so. Check this before anything else, because it decides whether a server is usable at all.

Vendor-backed servers and what they add

Eraser runs an MCP server so an MCP client can create and edit diagrams in an Eraser workspace, which is the right choice if your team already has that workspace and wants diagrams to land in it. Lucid Software publishes an official Lucid MCP server that creates diagrams and retrieves documents from Lucidchart and Lucidspark, using an OAuth flow rather than an API key. Both bind the diagram to a hosted account, which is a feature if that is where your diagrams live and a constraint if your code cannot leave the machine.

When this was checked

Every fact on this page was checked against the vendor's own site or repository on 22 August 2026. Prices and features change without notice, and a roundup is only as good as its last check, so verify anything you are about to make a decision on.

What to check before choosing

QuestionWhy it decides things
Does the model choose coordinates?If yes, expect overlaps past a few nodes
stdio or remote?Most clients accept stdio only
Where does the diagram live?Hosted workspace or a local file
Does it need an account?Blocks air-gapped and restricted use
What can you export?Decides whether it fits your docs
FAQ

Questions

Why do MCP-generated diagrams often come out overlapping?

Because most diagram servers ask the model for coordinates, and spatial arrangement is the least reliable thing a language model does. A server that accepts a semantic graph and runs a solver removes the failure rather than reducing it.

Can I use a remote MCP server?

It depends entirely on your client. Many accept stdio only, and a remote server that appears in a registry search is not necessarily connectable. Put a local stdio bridge in front of it if you need one.

Do I need an account to use one?

Vendor-backed servers generally bind to a hosted workspace and need one. A local server that writes files does not, which is what makes it usable in restricted environments.

Related

Last updated