FlexSim MCP Server
Control and automate FlexSim simulations through AI assistants like Claude, enabling manufacturing and warehouse digital twin analysis, parameter studies, and real-time model manipulation via tools for opening models, running simulations, evaluating FlexScript, and exporting results.
README
title: FlexSim MCP Server emoji: ⚙️ colorFrom: blue colorTo: green sdk: gradio sdk_version: "5.0.0" app_file: app.py pinned: false
FlexSim MCP Server (incl. FlexSimPy)
Control and automate FlexSim simulations via the Model Context Protocol (MCP). This repository exposes a FastMCP server that launches FlexSim (via FlexSimPy) and provides tools to open models, run/stop, step, query/set node values, evaluate FlexScript, export results, and more.
Why this matters
- Bridges digital twins and AI assistants (Claude/Cursor) for manufacturing/warehouse analysis and “what-if” studies.
- Demonstrates robust systems engineering: GUI process orchestration, async I/O, protocol handling, Windows quirks, and reproducible automation.
Design
┌─────────────────────┐ JSON-RPC ┌─────────────────────┐
│ MCP Client │◄────────────────────► │ MCP Server │
│ (Claude/Cursor) │ stdin/stdout │ (flexsim_mcp.py) │
└─────────────────────┘ └──────────┬──────────┘
│
│ Python API
▼
┌─────────────────────┐
│ FlexSimPy │
│ (Python SDK) │
└──────────┬──────────┘
│
│ COM/IPC
▼
┌─────────────────────┐
│ FlexSim │
│ (Simulation Engine)│
└─────────────────────┘
Features
- MCP server entry point:
mcp_server/flexsim_mcp.py - Tools exposed:
- Simulation control:
flexsim_open_model,flexsim_reset,flexsim_run,flexsim_run_to_time,flexsim_stop,flexsim_step,flexsim_get_time - Model/script:
flexsim_evaluate,flexsim_compile,flexsim_save_model,flexsim_new_model - Node access:
flexsim_get_node_value,flexsim_set_node_value - Results/stats:
flexsim_get_statistics,flexsim_export_results
- Simulation control:
- Logging:
mcp_server/flexsim_mcp.log - Reproducible environment:
uv sync - FlexSimPy build automation:
utility/build_automation.py - Integration tests:
tests/integration/test_mcp_client.py(stdio-based MCP client) - GUI sanity test:
utility/integration_test.py - Configurable behavior via
config.tomland environment overrides
Requirements
- OS: Windows
- Python: 3.12 (default; configurable via
config.tomlorFLEXSIM_PYTHON_VERSION) - Package manager:
uv(https://astral.sh) - FlexSim:
- Real runs require a local FlexSim installation (or repo-local mirror in
FlexSimDev/program) - FlexSimPy SDK submodule is provided at
depends/FlexSimPy(see.gitmodules)
- Real runs require a local FlexSim installation (or repo-local mirror in
Quickstart
Guided setup (recommended)
# From the repo root
python setup_mcp_server.py
# The setup helper will:
# - Summarize your configuration (install locations, Python version, submodule presence)
# - Ensure `uv` is installed and run `uv sync`
# - Build/install FlexSimPy via `utility/build_automation.py`
# - Optionally run a GUI integration sanity test
# - Print next steps to run the MCP server and tests
Manual setup
# 1) Install dependencies
uv sync
# 2) Check (and optionally build) FlexSimPy
uv run python utility/build_automation.py --status
# To force a build with a specific Python version:
# uv run python utility/build_automation.py --python-version 3.12
# 3) Launch MCP server in test mode (keeps the process alive; launches FlexSim on first tool call)
uv run mcp_server/flexsim_mcp.py --test-mode
# 4) Run integration tests (spawns the server as a subprocess; stdio MCP client)
uv run python tests/integration/test_mcp_client.py
Configuration
Primary configuration lives in config.toml (repo root). You can also override at runtime using environment variables.
Example (excerpt from config.toml):
[flexsim]
install_path = "FlexSimDev/program"
src_path = "C:\\Program Files\\FlexSim 2025 Update 2"
[python]
version = "3.12"
[server]
name = "FlexSimPy MCP Server"
version = "0.1.0"
http_endpoint = "http://127.0.0.1:8088/mcp"
[session]
reuse_policy = "singleton"
[logging]
level = "INFO"
log_file = "flexsim_mcp.log"
[build]
flexsimpy_dir = "depends/FlexSimPy"
auto_build = true
auto_install = true
Environment overrides (see utility/config.py):
FLEXSIM_CONFIG_PATH– path to an alternateconfig.tomlFLEXSIM_INSTALL_PATH– override FlexSim program pathFLEXSIM_PYTHON_VERSION– override Python version for FlexSimPyFLEXSIM_LOG_LEVEL– override logging level
Running the MCP server
Two modes:
- Test mode (interactive; no stdio protocol)
uv run mcp_server/flexsim_mcp.py --test-mode
# Keeps running; FlexSim launches on first tool call; intended for manual testing
- MCP stdio mode (default when no flags are passed)
uv run mcp_server/flexsim_mcp.py
# Expects MCP JSON-RPC messages via stdin/stdout; used by clients/tests
Client configuration (for Claude/Cursor/etc.):
{
"mcpServers": {
"flexsim": {
"command": "uv",
"args": [
"--directory",
"<repo-flexsim_mcp>/",
"run",
"mcp_server/flexsim_mcp.py"
]
}
}
}
Example tool calls (JSON-RPC via MCP)
Open a model
{"jsonrpc":"2.0","id":1,"method":"tools/call",
"params":{
"name":"flexsim_open_model",
"arguments":{"params":{
"model_path":"C:/path/to/Model.fsm"
}}
}}
Run to time and read time
{"jsonrpc":"2.0","id":2,"method":"tools/call",
"params":{"name":"flexsim_run_to_time","arguments":{"params":{"target_time":300}}}}
{"jsonrpc":"2.0","id":3,"method":"tools/call",
"params":{"name":"flexsim_get_time"}}
Evaluate FlexScript
{"jsonrpc":"2.0","id":4,"method":"tools/call",
"params":{"name":"flexsim_evaluate","arguments":{"params":{
"script":"getmodeltime()"
}}}}
Get/Set node value
{"jsonrpc":"2.0","id":5,"method":"tools/call",
"params":{"name":"flexsim_get_node_value","arguments":{"params":{
"node_path":"Model/Queue1/stats/input"
}}}}
{"jsonrpc":"2.0","id":6,"method":"tools/call",
"params":{"name":"flexsim_set_node_value","arguments":{"params":{
"node_path":"Model/Processor1/variables/processtime",
"value":5.0
}}}}
More FlexScript examples and behavior: see eval-flexscript.md.
Tests
Integration tests (stdio client; verifies tools and basic run sequence):
uv run python tests/integration/test_mcp_client.py
# Test 1: open model; Test 2: run for 20s, stop, verify time > 0
GUI sanity test (manual, interactive menu):
uv run python utility/integration_test.py
# Launches FlexSim GUI and provides simple interactive actions
Repository structure
mcp_server/flexsim_mcp.py— FastMCP server and FlexSimPy controller logic (entry point)utility/config.py— configuration loader with env overridesbuild_automation.py— FlexSimPy build/install orchestration (MSBuild resolution, .pyd install)copy_flexsim.py— copies a FlexSim installation intoFlexSimDev/(usesflexsim.src_path)integration_test.py— GUI-backed sanity test (suppresses CEF stderr, safe cleanup)utility.py— helpers: suppress stderr, kill FlexSim processes, force exit
tests/integration/test_mcp_client.py— async stdio client that starts the server and calls toolsREADME.md— test descriptions and sample outputs
depends/FlexSimPy— FlexSimPy SDK submoduleFlexSimDev/program— optional repo-local mirror of FlexSim installconfig.toml,mcp_server_config.json,AGENTS.md,eval-flexscript.md
Batch Experiment Orchestrator (Design)
A high-impact enhancement to run parameter sweeps with replications, gather metrics, and export a unified summary.csv (and manifest.json), enabling “what-if” studies at scale.
- Design document:
Batch_Experiment_Orchestrator.md - Status: design in-repo; implementation planned (not yet in
flexsim_mcp.py) - MVP tool (planned):
flexsim_run_experimentswith parameter list, seeds/replications, run_to_time, stats, artifact_dir
Troubleshooting
- FlexSim not found:
- Update
flexsim.install_pathinconfig.tomlor setFLEXSIM_INSTALL_PATH - Optionally use
utility/copy_flexsim.pyto mirror an existing install intoFlexSimDev/
- Update
- FlexSimPy unavailable:
- Check status:
uv run python utility/build_automation.py --status - Rebuild for the active Python version and ensure the
.pydis installed
- Check status:
- Stdio client issues:
- Ensure you run the server without
--test-modewhen using stdio clients - Check
mcp_server/flexsim_mcp.logfor details
- Ensure you run the server without
- GUI noise (CEF) or shutdown issues:
- The utilities suppress CEF stderr and force a safe interpreter exit when needed
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。