YADE MCP Server

yade-mcp

English | 简体中文

O.engines += [LLM()] # yet another engine.

yade-mcp connects AI agents to YADE — the open-source discrete element method engine — through the Model Context Protocol. Browse API docs, run simulations, and execute code, all through natural conversation.

Your agent doesn't just call tools — it sits at your YADE console, runs long simulations on its own, and stays in sync with what you're doing.

Works with any MCP client — verified with Claude Code, Codex CLI, Gemini CLI, GitHub Copilot CLI, OpenCode, and toyoura-nagisa.

Features

Your agent types, YADE runs

Powered by yade_execute_code

Describe what you want in plain language. The agent types commands into your YADE console — inspecting particles, tweaking parameters, stepping the engine, analyzing results. It reads each output, debugs, and iterates, the same way you do at the console yourself.

Set it running, walk away

Powered by yade_execute_task + yade_check_task_status + yade_interrupt_task

Run a full YADE script as a background task — just like firing off yade script.py, except you don't have to babysit it. The agent watches on its own: tailing the live output, catching errors as they appear, stopping the run gracefully when something looks off, fixing the script, and resubmitting — until the simulation actually finishes.

New session, no cold start

Powered by yade_list_tasks + yade_check_task_status

Every task you've submitted — the script, the live output, the final state — stays on record. When the context window fills up or you come back the next day, a fresh agent walks into a project that already remembers itself: it lists what's been run, reads what each task produced, and picks up without you re-explaining anything.

A live shell into the running simulation

Powered by yade_execute_code

While a task runs, the agent has a live shell into the simulation — ask it to inspect any variable, dump any object's state, or render a fresh plot on demand, without editing the script or stopping the run.

You type, the agent's in sync

Beyond submitted tasks, every line you type into the YADE console — the variables you peeked at, the parameters you tested, the dead ends you walked away from — flows into the agent's context too. When you turn to chat, it already has the trail of what you've been trying. Learning YADE and want feedback on what you just typed? Stuck on an unexpected error? Just ask — the agent saw what you typed and how YADE answered.

Tools (7)

Two documentation tools (no bridge) and five execution tools (bridge required):

Tool	Purpose	Bridge
yade_browse_api	Walk the YADE Python class tree	No
yade_query_api	BM25 keyword search across the API	No
yade_execute_code	Run Python in the live YADE process; returns synchronously	Yes
yade_execute_task	Submit a script as a long-running background task	Yes
yade_check_task_status	Inspect a running or finished task (output, status)	Yes
yade_interrupt_task	Gracefully stop a running task	Yes
yade_list_tasks	List submitted tasks with metadata	Yes

Quick Start

Prerequisites

• YADE installed
• uv installed (for uvx)

Agentic Setup (Recommended)

Copy this to your AI agent and let it self-configure:

Fetch and follow this bootstrap guide end-to-end:
https://raw.githubusercontent.com/yusong652/yade-mcp/master/docs/agentic/yade-mcp-bootstrap.md

Manual Setup

1. Register the MCP server in your client config:

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

2. Start the bridge inside YADE:

In a YADE Python console, install the bridge using YADE's own interpreter:

import sys, subprocess
subprocess.check_call([sys.executable, "-m", "pip", "install", "--user", "yade-mcp-bridge"])

On PEP 668 externally-managed environments (pip refuses --user), see the bootstrap guide for a portable form.

Restart YADE, then in the Python console:

import yade_mcp_bridge
yade_mcp_bridge.start()

Verify

Restart your AI agent (Claude Code, Codex CLI, Gemini CLI, etc.) and ask it to call yade_execute_code to verify the connection.

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT — see LICENSE.