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., "@materials-studio-mcp-moc Check if my structure is ready for Forcite and list missing parameters."
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.
这是一个面向 Windows 和 BIOVIA Materials Studio 2023/23.1 的 MCP 服务。它把 Materials Studio、Forcite、CASTEP 输入准备、地质模型构建和 LAMMPS/VMD 前处理封装成可审计的工具，同时保留严格的科学输入边界。
本项目的重点不是“自动猜参数”，而是让每一步都能回答三个问题：输入来自哪里、执行了什么、结果能否复现。
检测本机 Materials Studio 2023 安装并读取已安装的本地帮助文档。
扫描和解析 .xsd、.xtd、.stp 等结构文件，检查原子、键、晶胞和拓扑信息。
生成受控的超胞、表面、替换、羟基化、反离子和周期性水/盐模型候选。
在固定的 COMPASSIII、PCFF、Dreiding/QEq、Universal/QEq 配置下准备 Forcite 力场输入，并记录类型、键级、电荷和自动项检查结果。
生成受哈希约束的 CASTEP 独立输入候选；通用 CASTEP 计算执行和通用结果解析仍然关闭。
对模型规格做只读就绪度评估，并给出缺口补齐计划。
对 PubChem 化合物身份元数据和 Crossref 文献元数据提供受限、默认 dry-run 的公开证据查询。
将 CAR/MDF、LAMMPS data、VMD 检查和科学质量门组织为可追溯的项目流程。
通过异步任务提交、查询、取消和重试管理受控操作。
Related MCP server: comsol-codex-mcp
不接受任意 Perl、任意 MaterialsScript 或自然语言拼接脚本。
不自动编造晶体结构、晶胞、力场参数、交叉项、部分电荷或科学结论。
公开证据查询默认不联网；实时查询必须由用户明确授权并使用一次性确认。
不下载结构文件、力场文件、脚本或可执行文件作为“自动补齐”。
所有写入和计算工具都有 dry_run；真实的 R2/R3 操作需要精确、一次性的确认令牌。
通用 castep.calculation 和 results.castep_parsing 能力保持 unverified，不会因为计划或离线测试而开放。
资格计算只用于内部验证，不代表目标材料已经收敛，也不会自动获得生产许可。
典型的受控链路如下：
md_model_readiness_assess 读取模型规格，判断 ready、resolvable 或 blocked。
md_model_gap_resolution_plan 列出缺失结构、力场、晶胞、电荷和条件证据；缺失项由用户选择和审核。
md_project_initialize、md_project_update_specification 和 md_project_register_artifact 建立可复现项目记录。
md_structure_preflight、力场准备、CAR/MDF 导出和 LAMMPS 转换逐级检查哈希、计数、单位和质量门。
只有在目标模型证据充分、人工授权和对应工具边界都满足时，才进入受控执行；候选和资格结果不会自动变成生产结果。
在构建或计算新体系前，先提供已经确认的模型信息：组分和数量、相态、目标引擎、结构来源、力场状态、电荷方法、晶胞和边界条件。
可使用以下工具：
md_model_readiness_assess：只读检查已知字段和证据完整性。
md_model_gap_resolution_plan：生成有顺序、有边界的补齐计划，不代替科学决策。
md_search_public_model_evidence：只查固定的 PubChem/Crossref 元数据；默认 network_access=not_requested。
公开检索只产生“待人工复核的来源线索”，不会直接变成力场参数或结构输入。
在 Windows PowerShell 中运行：
.\install.ps1安装脚本会：
创建或修复 .venv；
安装锁定版本的 Python 依赖并执行 pip check；
创建 ASCII 路径运行别名；
生成本机 MCP 客户端配置 mcp-config.local.json。
也可以手动安装：
python -m venv .venv
.\.venv\Scripts\python.exe -m pip install --upgrade pip
.\.venv\Scripts\python.exe -m pip install -e .config/software.local.json、config/research-environment.local.json 和生成的 mcp-config.local.json 都是本机配置。里面的 Windows 路径只是示例，使用前必须根据本机安装位置修改；不要把个人凭据写入这些文件。
推荐使用部署目录的 ASCII 路径：
E:\ms_mcp\deployments\current\.venv\Scripts\python.exe -m materials_studio_mcp.server本地开发环境也可以使用：
.\.venv\Scripts\python.exe -m materials_studio_mcp.server命令行入口为：
.\.venv\Scripts\materials-studio-mcp.exe如果虚拟环境仍指向已经移动的 Python，请重新运行 .\install.ps1，并优先使用 ASCII 运行别名。
仓库中的 mcp-config.example.json 是模板。典型配置如下，请将路径替换为本机实际路径：
{
"mcpServers": {
"materials-studio-2023": {
"command": "E:\\ms_mcp\\deployments\\current\\.venv\\Scripts\\python.exe",
"args": ["-m", "materials_studio_mcp.server"],
"cwd": "E:\\ms_mcp\\deployments\\current",
"env": {
"MATERIALS_STUDIO_ROOT": "D:\\Program Files (x86)\\BIOVIA\\Materials Studio 23.1",
"MATERIALS_STUDIO_MCP_ROOT": "E:\\ms_mcp\\deployments\\current",
"MS_MOC_MCP_ROOT": "E:\\ms_mcp\\deployments\\current"
}
}
}
}公开工具数量和精确签名以 ms://catalog/public-tools 与 ms_task_catalog 为准。主要分层如下：
R0 只读检查：安装检测、本地帮助、工作区扫描、结构预检、项目读取、能力登记和模型就绪度。
R1 规划与证据：工作流建议、CASTEP 输入规划、公开证据 dry-run 和生产确认准备。
R2 受控写入：项目初始化、受哈希约束的结构构建、CAR/MDF 导出和 LAMMPS 转换。
R3 受控计算：仅开放经过固定 profile、人工授权和证据审计的 Forcite/资格流程；不提供通用 CASTEP MCP 执行接口。
config/ 本机软件、策略和科学合同配置
docs/validation/ 验收说明与哈希绑定回执
moc/ Materials Studio MOC 接口和桥接文件
scripts/ 安装、发布、候选验收和回滚脚本
src/materials_studio_mcp/ MCP 服务实现
tests/ 单元测试、合同测试和安全测试
install.ps1 本机安装入口
release-manifest.json 发布文件与哈希清单
requirements.lock 锁定依赖最近一次候选验收包含：
347 项源回归测试通过；
P6 模型就绪度与公开证据专项测试通过；
53 个公开工具登记一致；
能力登记的声明证据和有效证据哈希一致；
Windows Materials Studio Perl locale 子进程回退通过，stderr 为 0；
未启动 CASTEP、MPI 或 Materials Studio 受限计算进程。
版本以 release-manifest.json 为准。候选发布、切换和回滚脚本都会验证发布清单、依赖、哈希、locale 和 current 指针；候选验收不会自动切换生产部署。
运行完整回归：
.\.venv\Scripts\python.exe -m unittest discover -s tests -q
.\.venv\Scripts\python.exe -m pip check验证发布清单：
.\.venv\Scripts\python.exe -m materials_studio_mcp.release verify --manifest .\release-manifest.json如需构建候选发布，请使用 scripts/build_release_v1.ps1；如需安装候选，请使用 scripts/install_release_v1.ps1，不要直接修改不可变部署目录或 current Junction。
本仓库公开的是 MCP 服务源代码、测试、模板和审计回执。回执中的本机软件路径用于说明证据绑定，不是可供他人访问的文件共享路径；使用者应替换为自己的环境配置。
本项目不承诺自动完成任意材料体系的科学建模。它提供可复现的工具链、输入检查和证据管理，最终的结构、力场、晶胞、电荷和计算设置仍需材料研究者确认。
Audit the public MCP surface against the reviewed Windows/MS 2023 safety baseline.
Inputs: no named parameters. Returns: object
Generate an exact-count neutral quartz ClayFF/SPC/E/Joung LAMMPS candidate.
Inputs: max_atoms (required), output_slot (required), input_sha256 (required), idempotency_key (required), +16 more. Returns: object
Convert a hash-bound project CAR/MDF pair with a hash-bound msi2lmp forcefield.
Inputs: car_path (required), mdf_path (required), car_sha256 (required), mdf_sha256 (required), +9 more. Returns: object
Export a hash-bound project XSD to an audited CAR/MDF pair.
Inputs: input_xsd (required), output_slot (required), input_sha256 (required), idempotency_key (required), +4 more. Returns: object
Run only the bounded G01 MS -> LAMMPS -> VMD qualification workflow.
Inputs: input_xsd (required), project_id (required), dry_run, nvt_steps, +10 more. Returns: object
Plan local and human-reviewed remedies for model inputs that are not yet ready.
Inputs: model_spec (required), search_roots. Returns: object
Read-only intake assessment for an incomplete Materials Studio model specification.
Inputs: model_spec (required), search_roots. Returns: object
Validate the CAR/MDF pair, forcefield class, and explicit msi2lmp parameter file before conversion.
Inputs: car_path (required), mdf_path, forcefield_file, forcefield_class. Returns: object
Return a redacted pipeline configuration summary safe for MCP clients.
Inputs: no named parameters. Returns: object
Check MS, LAMMPS, MPI, VMD, conversion, and packing tools without running a simulation.
Inputs: run_version_probes. Returns: object
Issue a short-lived, single-use confirmation for an exact production operation.
Inputs: tool_name (required), parameters (required), ttl_seconds. Returns: object
Read a pipeline project's manifest.
Inputs: project_directory (required). Returns: object
Create a non-overwriting, reproducible MS/LAMMPS/VMD project directory and manifest.
Inputs: title (required), project_id (required), dry_run, projects_root. Returns: object
Register a project file with role, provenance, size, and SHA-256 integrity metadata.
Inputs: role (required), artifact_path (required), project_directory (required), source, +2 more. Returns: object
Record a quality-gate decision and structured evidence in the project manifest.
Inputs: gate (required), status (required), project_directory (required), dry_run, +1 more. Returns: object
Apply one audited, legal project lifecycle transition with a v1 result envelope.
Inputs: reason (required), target_status (required), project_directory (required), dry_run, +1 more. Returns: object
Replace the science contract and merge confirmed model/forcefield/geology requirements.
Inputs: specification (required), project_directory (required), dry_run, forcefield, +3 more. Returns: object
Validate project structure, model specification, forcefield declaration, and registered file integrity.
Inputs: project_directory (required). Returns: object
Audit a frozen target-model science contract without running any engine.
Inputs: project_directory (required), target_model_contract (required), dry_run, evidence_manifest, +1 more. Returns: object
Read fixed-provider PubChem/Crossref metadata only after explicit network opt-in.
Inputs: query (required), provider (required), dry_run, max_results, +2 more. Returns: object
Check an XSD or LAMMPS data file for structural, topology, type, cell, and charge issues.
Inputs: path (required), charge_tolerance. Returns: object
Plan or cancel one owned task and its fixed worker process tree.
Inputs: task_id (required), owner_capability (required), dry_run, confirmation_token. Returns: object
Query one persisted task after proving ownership.
Inputs: task_id (required), owner_capability (required). Returns: object
Preflight and retry one owned failed or cancelled task as a new immutable task.
Inputs: task_id (required), owner_capability (required), dry_run, confirmation_token. Returns: object
Preflight and asynchronously submit one exact governed operation.
Inputs: tool_name (required), parameters (required), dry_run, confirmation_token. Returns: object
Read-only preflight for the exact P3-C alpha-quartz CASTEP profile.
Inputs: input_manifest (required), input_manifest_sha256 (required). Returns: object
Inspect the local Gateway and report CASTEP submission blockers without submitting.
Inputs: requested_cores. Returns: object
Run one exact generated CASTEP PL through MatServer and exit before CASTEP.
Inputs: task_name (required), package_directory (required), package_manifest_sha256 (required), dry_run, +2 more. Returns: object
Detect the local BIOVIA Materials Studio 2023 installation and key paths.
Inputs: no named parameters. Returns: object
Plan a high-level workflow without executing it.
Natural-language requests are intentionally plan-only. Call the selected controlled tool explicitly after reviewing its typed parameters and gates.
Inputs: request (required), keep_job_dir, input_structure, input_trajectory, +2 more. Returns: object
Find local Materials Studio scripting code examples relevant to a query.
Inputs: query (required), max_results. Returns: object
Run one hash-bound Forcite profile and preserve its complete evidence bundle.
Inputs: profile_id (required), output_slot (required), input_sha256 (required), idempotency_key (required), +6 more. Returns: object
Protonate an explicit ledger of singly Si-coordinated O sites in a 2D p1 surface.
Inputs: sites (required), max_atoms (required), output_slot (required), input_sha256 (required), +10 more. Returns: object
Apply an explicit, named P1 substitution ledger and return a nonproduction candidate.
Inputs: output_slot (required), input_sha256 (required), substitutions (required), idempotency_key (required), +5 more. Returns: object
Read and fail-closed validate a mineral nanopore construction contract.
Inputs: contract_path (required). Returns: object
Convert a reviewed 2D surface to a hash-bound 3D periodic slab cell.
Inputs: max_atoms (required), output_slot (required), input_sha256 (required), idempotency_key (required), +8 more. Returns: object
Build one audited XSD supercell with the local MS 2023 BuildSuperCell API.
Inputs: repeat_a (required), repeat_b (required), repeat_c (required), max_atoms (required), +8 more. Returns: object
Enumerate nonproduction surface candidates without selecting or repairing a termination.
Inputs: miller_h (required), miller_k (required), miller_l (required), output_slot (required), +12 more. Returns: object
Import a hash-bound CIF/XSD crystal parent and verify its periodic element inventory.
Inputs: max_atoms (required), output_slot (required), input_sha256 (required), idempotency_key (required), +6 more. Returns: object
Place explicit counterions from a reviewed P1 fractional-coordinate ledger.
Inputs: max_atoms (required), placements (required), output_slot (required), input_sha256 (required), +8 more. Returns: object
Parse a Materials Studio document or project file and return best-effort metadata.
Inputs: path (required). Returns: object
List analysis targets through a hash-bound, confirmed MaterialsScript run.
Inputs: input_sha256 (required), input_document (required), dry_run, job_name, +3 more. Returns: object
List built-in Materials Studio example documents such as .xsd and .xtd.
Inputs: pattern, max_results. Returns: object
Return the readiness of the local Materials Studio desktop-control layer.
Inputs: no named parameters. Returns: object
Open one hash-bound project document through MOC after an optional dry run.
Inputs: document_path (required), document_sha256 (required), idempotency_key (required), project_directory (required), +2 more. Returns: object
Pack exact SPC/E water and NaCl counts into an audited periodic slab cell.
Inputs: ion_model (required), output_slot (required), random_seed (required), water_count (required), +14 more. Returns: object
Prepare hash-bound CASTEP PL/XSD folders, optionally bound to an adaptive plan.
Inputs: spins (required), input_xsd (required), input_sha256 (required), calculation_name (required), +17 more. Returns: object
Prepare hash-bound, execution-blocked .cell/.param files for standalone CASTEP.
Inputs: input_xsd (required), input_sha256 (required), calculation_name (required), output_directory (required), +3 more. Returns: object
Read a specific local Materials Studio help page and extract cleaned text and code examples.
Inputs: path (required). Returns: object
Recommend the best available Materials Studio MCP workflow for a natural-language task request.
Inputs: request (required), input_structure, calculation_context. Returns: object
Scan a workspace folder for Materials Studio files.
Inputs: root_dir (required), patterns, max_results. Returns: object
Search the installed Materials Studio 2023 scripting help.
Inputs: query (required), max_results. Returns: object
List the high-level workflows and helper tools exposed by this Materials Studio MCP server.
Inputs: no named parameters. Returns: object
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 53 tools
Many tools have overlapping purposes, such as multiple preflight/check/validate tools (md_structure_preflight, ms_castep_fixed_profile_preflight, ms_castep_preflight_checked, md_msi2lmp_preflight) and several prepare/build tools. The distinction between 'preflight', 'checked', and 'readiness' tools is not clear from names, causing potential misselection.
Tool names mix conventions: some use verb_noun (md_task_submit), others use noun_noun (ms_castep_gateway_readiness), and prefixes vary between ms_, md_, and combinations. Verbs like get, check, assess, plan, prepare, and run are used inconsistently, making the naming pattern unpredictable.
With 53 tools, the server is over 25, which is excessive for most MCP servers. While the domain is complex, many tools are highly specialized and could be consolidated (e.g., multiple preflight and checked variants), making the surface feel bloated.
The tool set covers the core lifecycle of Materials Studio workflows, including project initialization, structure preparation, conversion, validation, task management, and quality gates. Minor gaps exist, such as no generic list projects or delete artifact tools, but agents can work around these via existing project_get and register_artifact.
Hosted MCP server for task-first delegation to remote workstations and workers.
QuLab MCP remote server (Streamable HTTP) for computational science and lab tools.
MCP server for building and testing AI agents with multi-model experimentation and insights.