diagrams-mcp-server

MCP server for generating cloud architecture diagrams, flowcharts, sequence diagrams, and more — powered by three rendering engines: mingrammer/diagrams, Mermaid, and PlantUML.

Getting Started

Hosted (Recommended)

Connect to the public hosted server — no installation required. All rendering engines and dependencies are pre-installed.

Claude Desktop

Add to your `claude_desktop_config.json` (`Settings` → `Developer` → `Edit Config`):

{
  "mcpServers": {
    "diagrams-mcp": {
      "url": "https://diagrams-mcp-production.up.railway.app/mcp"
    }
  }
}

Claude Code (CLI)

Run:

claude mcp add diagrams-mcp https://diagrams-mcp-production.up.railway.app/mcp

Or add to your `.mcp.json`:

{
  "mcpServers": {
    "diagrams-mcp": {
      "url": "https://diagrams-mcp-production.up.railway.app/mcp"
    }
  }
}

Cursor

Add to your `.cursor/mcp.json`:

{
  "mcpServers": {
    "diagrams-mcp": {
      "url": "https://diagrams-mcp-production.up.railway.app/mcp"
    }
  }
}

Windsurf

Add to your `~/.codeium/windsurf/mcp_config.json`:

{
  "mcpServers": {
    "diagrams-mcp": {
      "serverUrl": "https://diagrams-mcp-production.up.railway.app/mcp"
    }
  }
}

VS Code

Add to your `.vscode/mcp.json`:

{
  "servers": {
    "diagrams-mcp": {
      "type": "http",
      "url": "https://diagrams-mcp-production.up.railway.app/mcp"
    }
  }
}

Local Installation

Prerequisites

Graphviz is required for the default local/in-process rendering mode. Mermaid CLI and PlantUML are optional — install them only if you need those specific rendering engines locally.

> Note: The hosted server runs as a slim MCP facade plus a separate renderer service, and has all render dependencies pre-installed in the renderer. Local prerequisites only apply if you're running in-process rendering yourself.

Install the server

Via uvx (recommended):

uvx diagrams-mcp-server

Via pip:

pip install diagrams-mcp-server

From source:

pip install git+https://github.com/ByteOverDev/diagrams-mcp.git

Configure your MCP client

Claude Desktop

Add to your `claude_desktop_config.json` (`Settings` → `Developer` → `Edit Config`):

uvx (recommended):

{
  "mcpServers": {
    "diagrams-mcp": {
      "command": "uvx",
      "args": ["diagrams-mcp-server"]
    }
  }
}

pip:

{
  "mcpServers": {
    "diagrams-mcp": {
      "command": "diagrams-mcp-server"
    }
  }
}

Claude Code (CLI)

Run:

claude mcp add diagrams-mcp -- uvx diagrams-mcp-server

Or add to your `.mcp.json`:

uvx (recommended):

{
  "mcpServers": {
    "diagrams-mcp": {
      "command": "uvx",
      "args": ["diagrams-mcp-server"]
    }
  }
}

pip:

{
  "mcpServers": {
    "diagrams-mcp": {
      "command": "diagrams-mcp-server"
    }
  }
}

Cursor

Add to your `.cursor/mcp.json`:

uvx (recommended):

{
  "mcpServers": {
    "diagrams-mcp": {
      "command": "uvx",
      "args": ["diagrams-mcp-server"]
    }
  }
}

pip:

{
  "mcpServers": {
    "diagrams-mcp": {
      "command": "diagrams-mcp-server"
    }
  }
}

Windsurf

Add to your `~/.codeium/windsurf/mcp_config.json`:

uvx (recommended):

{
  "mcpServers": {
    "diagrams-mcp": {
      "command": "uvx",
      "args": ["diagrams-mcp-server"]
    }
  }
}

pip:

{
  "mcpServers": {
    "diagrams-mcp": {
      "command": "diagrams-mcp-server"
    }
  }
}

VS Code

Add to your `.vscode/mcp.json`:

uvx (recommended):

{
  "servers": {
    "diagrams-mcp": {
      "type": "stdio",
      "command": "uvx",
      "args": ["diagrams-mcp-server"]
    }
  }
}

pip:

{
  "servers": {
    "diagrams-mcp": {
      "type": "stdio",
      "command": "diagrams-mcp-server"
    }
  }
}

Available Tools

Discovery

• `list_providers()` → `list[str]` — List all diagram providers (`aws`, `gcp`, `k8s`, `azure`, `onprem`, etc.)
• `list_services(provider)` → `list[str]` — List service categories within a provider (e.g. `aws` → `compute`, `database`, `network`)
• `list_nodes(provider, service)` → `list[dict]` — List node classes for a provider.service pair with import paths
• `search_nodes(query)` → `list[dict]` — Search for nodes by keyword across all providers (e.g. "postgres", "lambda")

Rendering

• `render_diagram(code)` → `Image` (PNG) — Execute a Python script using mingrammer/diagrams in a sandboxed subprocess. Returns a rendered cloud architecture diagram.
• `render_mermaid(definition)` → `Image` (PNG/SVG) — Render a Mermaid diagram definition (flowcharts, sequence, class, ER, state, Gantt, and more).
• `render_plantuml(definition)` → `Image` (PNG) — Render a PlantUML diagram definition (sequence, class, component, activity, state, deployment).

Cross-Provider Equivalence

• `find_equivalent(node, target_provider?)` → `dict` — Find equivalent services across cloud providers (e.g. `EC2` → `ComputeEngine` on GCP).
• `list_categories()` → `list[dict]` — List all 30 infrastructure role categories with mapped nodes across providers.

Resources

The server provides reference documentation accessible via MCP resource URIs:

Examples

Cloud Architecture (mingrammer/diagrams)

> "Draw an AWS architecture with an ALB routing to two ECS services, backed by RDS and ElastiCache"

from diagrams import Diagram, Cluster
from diagrams.aws.network import ALB
from diagrams.aws.compute import ECS
from diagrams.aws.database import RDS, ElastiCache

with Diagram("ECS Service", direction="LR"):
    lb = ALB("ALB")

    with Cluster("ECS Cluster"):
        services = [ECS("Web"), ECS("API")]

    lb >> services
    services[0] >> ElastiCache("Cache")
    services[1] >> RDS("Database")

Flowchart (Mermaid)

> "Create a flowchart showing a CI/CD pipeline"

Sequence Diagram (PlantUML)

> "Show the authentication flow between a client, API gateway, and auth service"

@startuml
Client -> "API Gateway": POST /login
"API Gateway" -> "Auth Service": Validate credentials
"Auth Service" --> "API Gateway": JWT token
"API Gateway" --> Client: 200 OK + token
Client -> "API Gateway": GET /data (Bearer token)
"API Gateway" -> "Auth Service": Verify token
"Auth Service" --> "API Gateway": Valid
"API Gateway" --> Client: 200 OK + data
@enduml

Development

# Clone and install
git clone https://github.com/ByteOverDev/diagrams-mcp.git
cd diagrams-mcp
pip install -e ".[dev]"

# Run tests
pytest

# Lint and format
ruff check .
ruff format .

# Run the MCP server locally (stdio mode)
diagrams-mcp-server

Split Facade/Renderer Mode

For hosted deployments, the MCP server can run as a lightweight facade that delegates render work to a separate renderer service. This keeps the always-on MCP process small while Graphviz, Chromium, Mermaid CLI, Java, and PlantUML live only in the renderer image.

# Terminal 1: renderer service
RENDERER_HOST=0.0.0.0 RENDERER_PORT=8001 diagrams-renderer-server

# Terminal 2: HTTP MCP facade delegating to the renderer
FASTMCP_TRANSPORT=http \
FASTMCP_HOST=0.0.0.0 \
FASTMCP_PORT=8000 \
DIAGRAMS_RENDERER_MODE=remote \
DIAGRAMS_RENDERER_URL=http://127.0.0.1:8001 \
diagrams-mcp-server

Docker/Railway examples are included:

Key environment variables:

Supported Providers

The `render_diagram` tool supports all providers from the mingrammer/diagrams library, including:

AWS, GCP, Azure, Kubernetes, On-Premise, AlibabaCloud, OCI, OpenStack, DigitalOcean, Elastic, Outscale, Generic, and Custom nodes.

Use `list_providers()` and `search_nodes(query)` to discover available nodes.

License

MIT