Excalidraw

Excalidraw is an open-source virtual whiteboard for hand-drawn style diagrams. Excalidraw does not publish an official hosted MCP endpoint, so you connect it through the community Excalidraw MCP server (mcp-excalidraw-server), which you self-host alongside a live canvas UI. Agents create and edit diagram elements through MCP tools while you watch the canvas update in real time.

Unlike Slack and GitHub, you do not create an OAuth app or provide a Client ID and Client Secret for Excalidraw. Like Zoho, you add the MCP server URL to QuilrAI manually.

See Overview for prerequisites and secret-handling guidance.

How Excalidraw Differs

• No OAuth credentials. The Excalidraw MCP server has no built-in authentication, so there is no Client ID or Client Secret to create or paste into QuilrAI. Access control comes from your network and from the gateway's access controls.
• Self-hosted. There is no vendor-hosted endpoint. You run the MCP server and its canvas yourself (Node.js or Docker) and expose it to the gateway.
• stdio transport by default. The server speaks MCP over stdio. QuilrAI registers MCPs by a transport URL ending in /sse or /mcp, so you put a stdio-to-HTTP bridge such as Supergateway in front of it.

What This MCP Can Do

The server exposes 26 tools:

Capability	Tools	Access
Create, read, update, and delete elements	create_element, get_element, update_element, delete_element, query_elements, batch_create_elements, duplicate_elements	Read and write
Arrange and organize elements	align_elements, distribute_elements, group_elements, ungroup_elements, lock_elements, unlock_elements	Write
Inspect the scene	describe_scene, get_canvas_screenshot	Read only
Import and export	export_scene, import_scene, export_to_image, export_to_excalidraw_url, create_from_mermaid	Read and write
Manage canvas state	clear_canvas, snapshot_scene, restore_snapshot	Write
Control the viewport	set_viewport	Write
Reference material	read_diagram_guide, get_resource	Read only

create_from_mermaid converts Mermaid diagram source into Excalidraw elements, which is the fastest way for an agent to draw architecture and flow diagrams. clear_canvas and restore_snapshot are destructive; consider disabling them with Tools Management if agents only need to draw.

Deploy The Excalidraw MCP Server

1. On the host that will run the whiteboard, start the canvas and MCP server. The npm package starts both; the canvas listens on port 3000 by default:

   npx -y mcp-excalidraw-server

   Docker images are also published: ghcr.io/yctimlin/mcp_excalidraw:latest (MCP server) and ghcr.io/yctimlin/mcp_excalidraw-canvas:latest (canvas).

2. Bridge the stdio MCP server to streamable HTTP with Supergateway:

   npx -y supergateway \
     --stdio "npx -y mcp-excalidraw-server" \
     --outputTransport streamableHttp \
     --port 8000

   The MCP endpoint is now http://<host>:8000/mcp.

3. Make the endpoint reachable from the QuilrAI gateway over HTTPS - for example behind your reverse proxy or an internal load balancer. Do not expose it to the public internet unprotected; the server has no authentication of its own.

4. In QuilrAI, go to the MCP Gateway tab, click Add MCP, and paste the /mcp URL manually. No OAuth authorization step is required.

5. Open the canvas UI (http://<host>:3000) in a browser, then ask a connected agent to draw something - for example "create a flowchart of our deployment pipeline" - and confirm the elements appear on the canvas in real time.

Configuration

Environment variable	Purpose
EXPRESS_SERVER_URL	Canvas server URL the MCP server syncs to. Default http://127.0.0.1:3000.
ENABLE_CANVAS_SYNC	Real-time sync of element changes to the canvas. Default true.
EXCALIDRAW_NO_AUTOSTART	Prevents the MCP server from auto-starting the canvas, for when the canvas runs as its own service or container.
EXCALIDRAW_EXPORT_DIR	Base directory for files written by the export tools.
PORT, HOST	Canvas bind address. The canvas binds to 127.0.0.1 by default.

Keep In Mind

• No built-in authentication. Anyone who can reach the bridged endpoint can drive the canvas. Keep it on a private network, terminate TLS in front of it, and restrict who can use the MCP with the gateway's access controls and agent configuration.
• The canvas is shared state. All connected agents and browsers see and edit the same scene. Use snapshot_scene before risky operations, and consider disabling clear_canvas for agent use.
• One canvas per server. To give teams isolated whiteboards, run separate instances (or containers) and add each one to QuilrAI as its own MCP.
• Exports land on the server host. export_to_image and export_scene write files under EXCALIDRAW_EXPORT_DIR on the machine running the MCP server, not on the agent's machine.

Troubleshooting

Error	Likely cause	Fix
Gateway cannot connect to the URL	The Supergateway bridge is not running, or the URL does not end in /mcp.	Confirm the bridge process is up, then re-copy the full endpoint URL including the /mcp path into QuilrAI.
Tools succeed but nothing appears on the canvas	The canvas server is not running, or the MCP server cannot reach it.	Start the canvas, and set EXPRESS_SERVER_URL to the canvas URL reachable from the MCP server. Check ENABLE_CANVAS_SYNC is not set to false.
Canvas only reachable from localhost	The canvas binds to 127.0.0.1 by default.	Set HOST=0.0.0.0 (or front it with a reverse proxy) and open the port to your users' network only.
Connection drops mid-session	The bridge ran in stateless mode and the session timed out.	Run Supergateway with --stateful and a suitable --sessionTimeout.

References

• Excalidraw MCP server (yctimlin/mcp_excalidraw)
• mcp-excalidraw-server on npm
• Supergateway: expose stdio MCP servers over HTTP
• Excalidraw