PlantUML MCP Server
# PlantUML MCP Server
[](https://lightnow.ai/servers/io.github.infobip/plantuml-mcp-server)
A [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) server that provides PlantUML diagram generation capabilities for Claude Desktop and Claude Code.
### Available Tools
1. **`generate_plantuml_diagram`** - Generate diagrams and get embeddable URLs (SVG/PNG), optionally save locally
2. **`encode_plantuml`** - Encode PlantUML code for URL sharing
3. **`decode_plantuml`** - Decode PlantUML from encoded strings
### Available Prompts
1. **`plantuml_error_handling`** - Guidelines for handling PlantUML syntax errors and implementing auto-fix workflows
This prompt provides Claude instances with comprehensive instructions on how to:
- Detect PlantUML syntax errors using native server validation
- Implement intelligent auto-fix workflows for common syntax issues
- Parse structured error responses and apply appropriate corrections
- Handle validation failures gracefully with retry logic
The prompt enables Claude to automatically detect and fix common PlantUML errors like missing tags, invalid arrow syntax, typos in keywords, and missing quotes, making PlantUML diagram generation more reliable and user-friendly.
## Quick Setup
### For Claude Code
```bash
# Using default PlantUML server
claude mcp add plantuml --scope user --env PLANTUML_SERVER_URL=https://www.plantuml.com/plantuml -- npx plantuml-mcp-server
```
### For Claude Desktop
Add this to your Claude Desktop MCP configuration (`~/Library/Application Support/Claude/claude_desktop_config.json` on macOS):
```json
{
"mcpServers": {
"plantuml": {
"command": "npx",
"args": ["plantuml-mcp-server"],
"env": {
"PLANTUML_SERVER_URL": "https://www.plantuml.com/plantuml"
}
}
}
}
```
To use your own PlantUML server, change the `PLANTUML_SERVER_URL` environment variable.
Then restart Claude Desktop/Code to activate the MCP server.
## What You Can Do
After setup, you can ask Claude to:
- **Generate PlantUML diagrams** and get embeddable SVG/PNG URLs
- **Create sequence diagrams, class diagrams, architecture diagrams**
- **Use advanced PlantUML features** like `!include` directives and external libraries
- **Encode/decode PlantUML** for URL sharing
- **Save diagrams locally** with secure path restrictions
### Saving Diagrams Locally
The `generate_plantuml_diagram` tool supports saving diagrams to local files via the `output_path` parameter.
**Security:** By default, files can only be saved within the current working directory. Only `.svg` and `.png` extensions are allowed.
To allow additional directories, set the `PLANTUML_ALLOWED_DIRS` environment variable:
```bash
# Allow specific directories (colon-separated)
PLANTUML_ALLOWED_DIRS=/home/user/diagrams:/tmp/output
# Allow any directory (use with caution)
PLANTUML_ALLOWED_DIRS=*
```
---
## Examples
### C4 diagram for plantuml-mcp-server
```
> add c4 diagram for this project in readme 'C4 diagram for plantuml-mcp-server' section
```

### Sequence diagram for plantuml-mcp-server
```
> add sequence diagram in readme in architecture section
```

### OAuth2 Authorization Code Flow Diagram
```
> show me basic oauth2 flow png, open it in browser
```

---
## Development Setup
### Prerequisites
- Node.js 22.12.0 or later
- npm
### Local Installation
```bash
# Clone and setup
git clone https://github.com/mzagar/plantuml-mcp-server.git
cd plantuml-mcp-server
make setup
```
### Development Commands
```bash
# Show all available commands
make help
# Build and run all tests
make clean build test-all
```
### Local MCP Smoke Test
Test the compiled server in this checkout through [`mcpc`](https://github.com/apify/mcpc):
Requires Node.js 22.12.0 or later, npm, Make, and network access.
```bash
# Install project dependencies, then build and test the local server
npm ci --ignore-scripts --no-audit --no-fund --engine-strict
make test-mcpc
```
If `mcpc` is missing from `PATH`, the target installs the official Apify CLI with `npm install --global --no-audit --no-fund @apify/mcpc`. It reuses an existing installation without upgrading it and checks `mcpc --version` before testing. The smoke test has been verified with mcpc 0.6.0.
This is a global CLI installation, not a project dependency. It requires permission to write to the npm global prefix. The target never uses `sudo`. If installation fails or the CLI is still not on `PATH`, the test stops with instructions. You can also install it manually with `npm install -g @apify/mcpc`.
`make test-mcp` is an alias. `make test-all` runs the unit/type checks and this MCP test, including the CLI installation check. Running `node scripts/test-mcpc.mjs` directly does not install the CLI.
The test verifies all three tools, the prompt, Unicode roundtrip, diagram URLs, SVG/PNG files, C4 rendering, syntax errors, and output-path restrictions.
It always renders synthetic diagrams through `https://www.plantuml.com/plantuml`, so network access is required. It does not use a published npm server or an existing user session.
The runner creates its own temporary config, private `MCPC_HOME_DIR`, working directory, and output directory. Its `@local` session is isolated from user sessions and closed after success, failure, or interruption. It retains images and `results.json` in the artifact directory printed by the command.
Success ends with `PASS local mcpc smoke test` and exit code 0. Missing prerequisites, network errors, failed assertions, or cleanup failures return a nonzero exit code.
For a trusted corporate CA, run:
```bash
NODE_EXTRA_CA_CERTS=/absolute/path/to/trusted-ca.pem make test-mcpc
```
The runner forwards this setting to the server. Do not disable TLS validation. If cleanup fails, use its printed private home: `MCPC_HOME_DIR="<artifact-directory>/mcpc" mcpc close @local`. Do not close sessions in your default mcpc home.
Agent instructions are in [AGENTS.md](AGENTS.md).
### Environment Variables
| Variable | Description | Default |
|----------|-------------|---------|
| `PLANTUML_SERVER_URL` | PlantUML server URL | `https://www.plantuml.com/plantuml` |
| `PLANTUML_ALLOWED_DIRS` | Additional directories for `output_path` (colon-separated, or `*` for any) | CWD only |
## License
MIT License - see [LICENSE](LICENSE) file for details.
## MCP Registry Listings
- [LightNow](https://lightnow.ai/servers/io.github.infobip/plantuml-mcp-server)
- [Smithery](https://smithery.ai/servers/infobip/plantuml-mcp-server)TDQS
Scored across 3 tools
Each tool has a clearly distinct purpose: decode_plantuml reverses encoding, encode_plantuml prepares code for URLs, and generate_plantuml_diagram creates diagrams with validation. There is no overlap or ambiguity between these functions, making tool selection straightforward for an agent.
All tool names follow a consistent verb_noun pattern with snake_case (decode_plantuml, encode_plantuml, generate_plantuml_diagram). The naming is predictable and readable, adhering to a uniform style throughout the set.
With 3 tools, this server is well-scoped for handling PlantUML encoding, decoding, and diagram generation. Each tool earns its place by covering essential operations without bloat, making the count appropriate for the domain.
The tool set provides complete coverage for core PlantUML workflows: encoding for URLs, decoding back to code, and generating diagrams with validation and error handling. There are no obvious gaps, as these tools support the full lifecycle from code to visual output.