Composer
Architecture Diagrams automated by your AI Agent
Composer MCP Server
Give your AI coding agent an architecture canvas.
Design, visualize, and evolve software architecture diagrams - right from your IDE.
Composer is a visual system design tool that lets AI coding agents create and modify interactive architecture diagrams through MCP (Model Context Protocol). Your agent gets tools to add services, databases, queues, and connections, and you get a live canvas at usecomposer.com that updates in real-time.
Your IDE <--> MCP Server (this package) <--> Composer API <--> Your Diagram
Getting Started
Connect your IDE
Claude Code:
claude mcp add --transport http composer https://mcp.usecomposer.com
Cursor — create .cursor/mcp.json in your project root:
{
"mcpServers": {
"composer": {
"type": "http",
"url": "https://mcp.usecomposer.com"
}
}
}
Your browser will open for authorization on first use.
Claude Code
claude mcp add --transport http composer https://mcp.usecomposer.com
Cursor
Create .cursor/mcp.json in your project root:
{
"mcpServers": {
"composer": {
"type": "http",
"url": "https://mcp.usecomposer.com"
}
}
}
Codex
codex mcp add composer -- npx -y @usecomposer/mcp --stdio
VS Code (Copilot Chat)
Create .vscode/mcp.json in your project root:
{
"servers": {
"composer": {
"type": "http",
"url": "https://mcp.usecomposer.com"
}
}
}
VS Code (Cline)
Open Cline sidebar > Settings (gear icon) > MCP Servers > Add Remote Server:
{
"mcpServers": {
"composer": {
"type": "http",
"url": "https://mcp.usecomposer.com"
}
}
}
VS Code (Continue)
Add to .continue/config.yaml:
mcpServers:
- name: composer
url: https://mcp.usecomposer.com
Windsurf
Add to ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"composer": {
"serverUrl": "https://mcp.usecomposer.com"
}
}
}
Note: Windsurf uses
"serverUrl"instead of"url".
OpenCode
Create opencode.json in your project root:
{
"mcp": {
"composer": {
"type": "remote",
"url": "https://mcp.usecomposer.com"
}
}
}
Tools
Diagram Management
| Tool | Description |
|---|---|
list_diagrams | List all your diagrams. Call this first to find which diagram to work on |
create_diagram | Create a new diagram and auto-select it for this session |
select_diagram | Select which diagram to work on for this session |
rename_diagram | Rename the currently selected diagram |
Note: Call
list_diagramsthenselect_diagram(orcreate_diagram) before using other tools.
Read
| Tool | Description |
|---|---|
get_graph | Get the full architecture diagram - all nodes and edges |
get_node | Get details for a single node including connected edges |
search_graph | Search nodes and edges by keyword |
get_screenshot | Get a PNG thumbnail of the diagram from the last auto-save |
Write
| Tool | Description |
|---|---|
upsert_node | Create or update a node (service, database, queue, etc.) |
upsert_edge | Create or update a connection between two nodes |
define_api | Define API endpoints on a backend service node |
delete_element | Delete a node or edge from the diagram |
link_path | Link a node to a file or folder in your codebase |
Plan & Verify
| Tool | Description |
|---|---|
verify_diagram | Check for issues like missing endpoints or orphaned nodes |
plan_import | Step-by-step workflow for importing an existing codebase |
get_guide | Reference guide for node types, protocols, and best practices |
Node Types
client · frontend · backend · database · cache · queue · storage · external
Edge Protocols
REST · gRPC · GraphQL · WebSocket · TCP · UDP · async · event · internal
Example Prompts
Once connected, try asking your AI agent:
| Prompt | What it does |
|---|---|
| "Import this codebase into Composer" | Scans your repo and builds the architecture diagram |
| "Create a new Composer diagram called Backend Architecture" | Creates and auto-selects a new diagram |
| "Add a Redis cache between the API and the database in Composer" | Adds a new node and edges to the diagram |
| "Add analytics monitoring to the Composer diagram" | Adds observability nodes and connections |
| "Update the APIs I defined in Composer" | Refreshes endpoint definitions on backend nodes |
| "Verify my Composer diagram" | Checks for missing endpoints, orphaned nodes, etc. |
| "Link each Composer node to its source code" | Associates diagram nodes with file paths |
How It Works
Authentication is handled via OAuth 2.1 — your browser opens for a one-time consent flow, and you're connected. Tool calls are proxied to the Composer API at mcp.usecomposer.com. Your diagram data lives on Composer's servers. The MCP server itself is stateless.
Changes made through MCP are immediately visible on the Composer canvas via real-time WebSocket sync.
Development
git clone https://github.com/olivergrabner/composer-mcp
cd composer-mcp
npm install
npm run dev # Watch mode (rebuilds on change)
npm run build # Production build
Links
- Composer - the visual architecture canvas
- MCP Protocol - Model Context Protocol spec
- Issues - bug reports and feature requests
License
MIT
Related Servers
Alpha Vantage MCP Server
sponsorAccess financial market data: realtime & historical stock, ETF, options, forex, crypto, commodities, fundamentals, technical indicators, & more
Flutter MCP
A real-time MCP server providing Flutter/Dart documentation and pub.dev package information to AI assistants, supporting over 50,000 packages on demand.
Overture
Visual plan approval for AI coding agents. See your agent's plan as an interactive graph, attach context, choose approaches, then approve before any code is written.
Apifox MCP Server
Provides API documentation from Apifox projects as a data source for AI programming tools that support MCP.
MCP Music Analysis
Analyze audio from local files, YouTube, or direct links using librosa.
Kite Trading MCP Server
An MCP server for the Zerodha Kite Connect API, featuring fully automated authentication without manual token handling.
MasterGo Magic MCP
A standalone MCP service that connects MasterGo design tools with AI models, enabling them to retrieve DSL data directly from design files.
Bring Your AI
Local-first Claude Code to Codex harness mover with a no-data remote MCP discovery surface.
DeepRank
Optimize any site for AI search: get DeepRank methodology, optimization steps, and suggestions (llms.txt, JSON-LD, audit checklist) so your AI assistant can implement AI visibility in the repo.
Jira Context MCP
MCP server to provide Jira Tickets information to AI coding agents like Cursor.
Neovim LSP MCP Server
Bridges AI coding assistants with Neovim's Language Server Protocol for AI-powered code intelligence and navigation.