wzrdbrain MCP Server
Click on "Deploy 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., "@wzrdbrain MCP Server generate a smooth 5-trick skating combo"
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.
wzrdbrain MCP Server
An MCP (Model Context Protocol) server that provides a physics-aware combo generator for wizard-style inline skating, powered by the wzrdbrain Python package.
Unlike random trick generators, this server understands skating physics. It models moves as state transitions (direction, edge, stance, weight point), ensuring that the combinations it generates are actually executable and flow naturally.
Prerequisites
Python 3.13+
uv for dependency management
Node.js (optional, for MCP Inspector)
Related MCP server: ca-scene-mcp
Installation
git clone https://github.com/your-username/wzrdbrain-mcp.git
cd wzrdbrain-mcp
uv syncClient Configuration
Claude Desktop
Open the configuration file for your operating system:
macOS:
~/Library/Application Support/Claude/claude_desktop_config.jsonWindows:
%APPDATA%\Claude\claude_desktop_config.json
Add the following, replacing <ABSOLUTE_PATH_TO_PROJECT> with the actual path:
{
"mcpServers": {
"wzrdbrain": {
"command": "uv",
"args": [
"--directory",
"<ABSOLUTE_PATH_TO_PROJECT>",
"run",
"mcp-server-wzrdbrain"
]
}
}
}Restart Claude Desktop. The wzrdbrain tools will appear via the hammer icon in your chat.
Claude Code
claude mcp add wzrdbrain -- uv --directory <ABSOLUTE_PATH_TO_PROJECT> run mcp-server-wzrdbrainGemini CLI
Add to .gemini/settings.json (project-level) or ~/.gemini/settings.json (user-level), replacing <ABSOLUTE_PATH_TO_PROJECT> with the actual path:
{
"mcpServers": {
"wzrdbrain": {
"command": "uv",
"args": [
"--directory",
"<ABSOLUTE_PATH_TO_PROJECT>",
"run",
"mcp-server-wzrdbrain"
]
}
}
}See the Gemini CLI MCP docs for additional options like timeout and env.
MCP Inspector
Test the tools locally before connecting to a client:
npx @modelcontextprotocol/inspector uv run mcp-server-wzrdbrainTools
generate_skating_combo
Generates a physics-aware sequence of inline skating tricks.
Parameter | Type | Default | Range |
| integer | 3 | 1–20 |
Each line shows the trick name and its entry/exit state (direction/edge/stance/point):
1. Front Soul Slide: front/outside/open/all → front/outside/open/all
2. Front Mizu Slide: front/inside/open/all → front/inside/open/all
3. Front Fast Slide: front/outside/open/all → front/outside/open/alllist_trick_categories
Returns the available trick categories as a sorted list. No parameters.
['base', 'manual', 'pivot', 'slide', 'swivel', 'transition', 'turn']get_tricks_by_category
Lists all tricks in a given category. Case-insensitive.
Parameter | Type | Required |
| string | yes |
Tricks in category:
- Back Predator (Open)
- Back Predator One
- Front Predator (Open)
- Front Predator OneInvalid categories return an error listing valid options:
Error: Invalid category. Valid categories are: base, manual, pivot, slide, swivel, transition, turnPrompts
skating_practice_routine
A prompt template that instructs the LLM to generate a 4-trick combo and structure it into a 30-minute practice session:
Warm-up (5 minutes)
Trick Breakdown & Practice (15 minutes)
Combo Execution (5 minutes)
Cool-down (5 minutes)
Error Messages
Message | Cause |
| Passed a string, float, or non-integer type |
| Value outside the allowed range |
| Unrecognized category name |
| Unexpected exception from wzrdbrain |
Development
Setup
uv sync --extra testRunning Tests
# All tests
uv run pytest -v
# Single test
uv run pytest tests/test_server.py::TestGenerateSkatingComboHappy::test_default_three_tricks
# Coverage (target: 100%)
uv run pytest --cov=wzrdbrain_mcp --cov-report=term-missingProject Structure
src/wzrdbrain_mcp/server.py # All MCP tools, prompts, and the entry point
tests/test_server.py # Full test suite
pyproject.toml # Dependencies (pinned with ==) and pytest configArchitecture
FastMCP decorated functions are plain callables — tests call them directly without an MCP client.
wzrdbrainlibrary provides two touchpoints:wzrdbrain.generate_combo()for combo generation andMOVESdict for the trick catalog.State transitions are modeled as
direction/edge/stance/point.Input validation returns error strings, not exceptions. Maintain this pattern for new tools.
Output sanitization validates the structure of
wzrdbrainresponses before formatting.Logging goes to stderr only. User input is sanitized via
_sanitize_for_log()before logging to prevent log injection.Dependencies are pinned with exact versions (
==) inpyproject.tomland locked viauv.lock.
Adding a New Tool
Add a
@mcp.tool()function inserver.pyValidate all inputs (type checks, range bounds) — return error strings for invalid input
Sanitize outputs from
wzrdbrainbefore returningLog user-supplied values through
_sanitize_for_log()onlyAdd tests in
tests/test_server.py— direct calls for happy paths,unittest.mock.patchfor error pathsVerify coverage stays at 100%:
uv run pytest --cov=wzrdbrain_mcp --cov-report=term-missingUpdate this README with the new tool's documentation
Available Tools
3 toolsgenerate_skating_comboA
Generates a physics-aware sequence of wizard-style inline skating tricks.
Args: num_tricks: The number of tricks to chain together in the combo. Must be between 1 and 20. Defaults to 3.
Inputs: num_tricks. Returns: result
get_tricks_by_categoryA
Retrieves all tricks that fall under a specific category.
Args: category: The category name (e.g., 'pivot', 'slide'). Validated against known categories.
Inputs: category (required). Returns: result
list_trick_categoriesA
Retrieves the available categories of inline skating tricks.
Inputs: no named parameters. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 3 tools
Each tool has a distinct purpose: listing categories, retrieving tricks by category, and generating combos. There is no meaningful overlap or risk of selecting the wrong tool.
All names use snake_case and follow a clear verb_noun style, but 'list_trick_categories' and 'get_tricks_by_category' mix two different retrieval verbs (list vs get), which is a minor inconsistency.
Three tools is well-suited to this narrow, focused domain. Each tool serves a distinct and necessary role without unnecessary bloat.
The core workflow—discover categories, retrieve tricks by category, and generate combos—is fully covered. Minor gaps like listing all tricks without a category or viewing individual trick details are workable but not essential.
Maintenance
Related MCP Connectors
Your own closet in your AI: outfits from clothes you own, rendered on a model inline.
Editing references, adaptable motion and brand memory for agents that edit video.
Generate cinematic images and video and chain shots into continuous motion.
Generate 3D models, worlds, PBR materials, and coordinated asset packs from AI agents.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceExposes MuJoCo physics simulation to AI assistants via 65 MCP tools, enabling natural language control of robotics simulation, trajectory optimization, contact analysis, and video export.12MIT
- AlicenseBqualityDmaintenanceEnables prompt-driven scene generation for computer animation using a JSON-serializable scene model and physics simulation backend.25Apache 2.0
- AlicenseNot gradedqualityBmaintenanceConnects AI agents to game engines at 60 FPS, enabling deterministic physics simulation, AI-to-rig animation synthesis, and safe sandboxed physics state control through the Model Context Protocol.Apache 2.0
- AlicenseNot gradedqualityBmaintenanceEnables multiple AI agents to each drive one wheeled robot inside a single shared warehouse physics world, with a coordinator assigning crates and stepping the lockstep simulation while robot connections submit motion commands through six to seven exposed tools. Supports deterministic replay, hash-chain verification, and offline headless rendering of first-person, chase, overhead, and combined camera views from the recorded run.MIT