inkwire
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@inkwireCreate a board so I can sketch the system architecture, then infer its structure."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
inkwire
A local MCP server with a shared drawing canvas, for collaborating visually with an AI agent on a codebase. You draw a system freehand in the browser; the server infers structure; Claude reads the board as data over MCP, edits it, and discusses it. Both of you write to the same board.
Quick start
yarn install
yarn buildInkwire ships as a Claude Code plugin: the MCP server, a Stop hook, the /use-inkwire and /back-to-claude-code commands, and the model-invocable trace-path skill. Install it once; it then loads in every session:
claude plugin marketplace add /path/to/inkwire
claude plugin install inkwire@inkwireAdjust CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS to 0 in your ~/.claude/settings.json
For a one-off run without installing, claude --plugin-dir /path/to/inkwire does the same for that session. Tool names carry the plugin prefix: mcp__plugin_inkwire_inkwire__boards_create.
The Session tab (below) needs two settings. Put them in ~/.claude/settings.json to make them permanent, or in the .claude/settings.json of the project you use inkwire from:
{
"env": { "CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS": "0" },
"permissions": { "defaultMode": "auto" }
}The canvas works without them. This repo's own .claude/settings.json already carries both.
Then ask Claude to create a board (boards_create). The tool result contains the panel URL — open it in your browser:
http://127.0.0.1:4691/?board=<board id>Draw with the pen (P), then press infer_structure (or ask Claude to run it). Closed shapes become nodes; connecting lines become edges. Claude renames the nodes after reading a screenshot.
Related MCP server: excalidraw-mcp
Session tab: talking in the panel
Type /use-inkwire in the terminal. Claude flips a server-held mode flag, and from then on delivers replies through the blocking session_send tool into the panel's SESSION tab, where you answer from the composer. A reply can carry a highlight: node and edge ids the canvas lights up. It can also carry a path: an ordered walk over a layer's edges, one caption per hop, that opens the scrubber across the top of the canvas and plays. Hold a layer chip to peek at its first path; click the chip's ▸ to keep the scrubber open and drag through it. The /trace-path skill writes one when you ask "walk me through" or "what happens when". /back-to-claude-code (typed, or the button in the strip) brings replies back to the terminal.
Two requirements, both checked by the server when the mode goes on, and both covered by the settings above:
Permission mode
autoorbypassPermissions— nobody is at the terminal to approve prompts. Switching modes inside the session works too; the check runs when/use-inkwiredoes.CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS=0— otherwise Claude Code moves the blocking call to a background task after two minutes.
The mode is not persisted; a server restart returns to the terminal. A session_send that waits 20 minutes with no answer returns idle and flips the mode back.
Development
yarn dev # server (tsx watch) + panel bundle (esbuild watch)
yarn test # full vitest suite
yarn typecheckConfiguration (env vars): INKWIRE_PORT (default 4691), INKWIRE_DATA_DIR (default ~/.inkwire — SQLite plus an images/ directory), INKWIRE_PROJECT_ROOT (the root that canvas_bind_code refs resolve against).
The design handoff that specifies this project lives in design_handoff_inkwire/. See CLAUDE.md for architecture notes.
License
MIT
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Connectors
AI agent draws editable hand-drawn diagrams (flowchart, ER, architecture) via MCP, not static images
Real-time collaborative whiteboard — AI agents and humans edit the same board live over MCP.
Intelligent product canvas: iterate on your product's design with your coding agent, synced to code.
Create and edit architecture diagrams from your AI agent; get an SVG and a live editable canvas.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceA shared whiteboard for you and your AI agent I wanted my AI agent and me to be able to point at the same thing. Any MCP-capable agent can read the canvas, draw on it, drop thought bubbles, animate elements, and react when you sketch something.1MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to create and manipulate live visual diagrams on an Excalidraw canvas in real-time via MCP tools.2,78314BSD 3-Clause
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to collaboratively draw and annotate Excalidraw diagrams in real-time via MCP tools, synced to a browser canvas.6Apache 2.0
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to draw animated, hand-drawn diagrams on a live whiteboard through MCP tools, allowing users to visually explain concepts and edit scenes interactively.45AGPL 3.0
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/TheTechChild/inkwire'
If you have feedback or need assistance with the MCP directory API, please join our Discord server