OpenSTAAD MCP Server
OfficialThis MCP server lets AI agents interact locally with Bentley STAAD.Pro models through the OpenSTAAD API.
Discover API guidance and load domain-specific skills before using tools.
List running STAAD.Pro instances, including aliases, PIDs, open models, and versions.
Check connection status, STAAD version, model path, analysis state, and allowed directories.
Execute sandboxed Python against a selected STAAD.Pro instance using the pre-connected
staadobject.Read CSV/XLSX input data into the sandbox and write results back to CSV/XLSX to avoid large context transfers.
Support multi-instance workflows by targeting specific STAAD.Pro instances.
Run locally with no cloud processing or telemetry, using stdio or HTTP transport modes.
Allows AI agents to interact with Bentley STAAD.Pro models for tasks like load case definition, data extraction, and property setting via the OpenSTAAD API.
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., "@OpenSTAAD MCP Server Set the steel grade of all beams to A992."
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.
OpenSTAAD MCP Server
A Model Context Protocol (MCP) server for Bentley STAAD.Pro that enables AI agents like Claude Desktop, Gemini, or VSCode Copilot to interact with your STAAD.Pro models and perform various time-consuming tasks like load cases definition, data extraction, repetitive property setting and more.
This MCP server was introduced as part of Bentley's Infrastructure AI Co-Innovation Initiative to help our users and accounts discover opportunities and innovate faster, while connecting Bentley's unique engineering tool capabilities to their emerging agentic workflows.
Key Features
Fast and flexible: Enjoy minimal latency, interact with every STAAD.Pro features covered by the OpenSTAAD API.
AI-friendly: Provides documentation, guidance and feedback via dedicated tools to help your AI agent ramp up quickly on the STAAD.Pro API.
Multi-instance support: Connects to multiple running STAAD.Pro instances simultaneously to parallelize tasks across models.
Privacy-first: All processing happens locally on your machine. No data is sent to the cloud. No telemetry.
Related MCP server: Qiao-MCP
Prerequisites
OS: Windows 11 or newer
STAAD.Pro 2025 or newer installed and running
Quick Start with Claude Desktop (<2min)
Download the latest
openstaad-mcp.mcpbfile from the GitHub Releases page.Open Claude Desktop.
Click the ☰ menu (top-left) → File → Settings → Extensions.
Click Advanced → Install Extensions.
Select the downloaded
.mcpbfile.Click the ☰ menu (top-left) → File → Exit
Restart Claude Desktop.
Claude Desktop will install the server automatically. Open a new conversation and ask Claude to interact with your STAAD.Pro model.
Tip: Make sure STAAD.Pro is running with a model open before you start chatting.
Other Clients & Configuration
TL;DR:
If not already installed, install uv with the command:
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"Configure your client to start the server in stdio mode with the command:
uvx --from git+https://github.com/BentleySystems/openstaad-mcp openstaad-mcpVS Code with GitHub Copilot
For stdio: Open the Command Palette → MCP: Add Server... → Command (stdio) and enter the following command:
uvx --from git+https://github.com/BentleySystems/openstaad-mcp openstaad-mcpFor http: First, start the server in a terminal:
uvx --from git+https://github.com/BentleySystems/openstaad-mcp openstaad-mcp --transport httpLook for the generated token and URL in the terminal output. It should look like this:
WARNING: No --token provided. Auto-generated token: abc123def456ghi789jkl012mno345pq INFO: Starting MCP server 'OpenSTAAD MCP' with transport 'http' (stateless) on http://127.0.0.1:18120/mcpThen, in VS Code, open the Command Palette → MCP: Add Server... → HTTP URL and enter the URL shown in the terminal (e.g.
http://127.0.0.1:18120/mcp).18120is the default port, but yours may differ if you have multiple instances running or if you changed the default. Add the headerAuthorization: Bearer <token>with the token shown in the MCP server terminal.
GitHub Copilot CLI
Use the /mcp add command inside a Copilot CLI session to add the server. See the Copilot CLI documentation for more details.
For stdio transport, use the command:
uvx --from git+https://github.com/BentleySystems/openstaad-mcp openstaad-mcpFor HTTP transport, first start the server in a terminal:
uvx --from git+https://github.com/BentleySystems/openstaad-mcp openstaad-mcp --transport httpLook for the generated token and URL in the terminal output. It should look like this:
WARNING: No --token provided. Auto-generated token: abc123def456ghi789jkl012mno345pq INFO: Starting MCP server 'OpenSTAAD MCP' with transport 'http' (stateless) on http://127.0.0.1:18120/mcpThen add the server in Copilot CLI using the URL shown in the terminal (e.g.
http://127.0.0.1:18120/mcp).18120is the default port, but yours may differ if you have multiple instances running or if you changed the default. Add the headerAuthorization: Bearer <token>with the token shown in the MCP server terminal.
Claude Desktop (manual configuration)
If you prefer manual setup over the .mcpb bundle, edit the Claude Desktop
config file directly:
Windows (MSIX):
%LOCALAPPDATA%\Packages\Claude_<id>\LocalCache\Roaming\Claude\claude_desktop_config.jsonWindows (classic):
%APPDATA%\Claude\claude_desktop_config.jsonmacOS:
~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"openstaad": {
"command": "uvx",
"args": ["--from", "git+https://github.com/BentleySystems/openstaad-mcp", "openstaad-mcp"]
}
}
}Claude Code (CLI)
For stdio transport, use the command:
claude mcp add --transport stdio openstaad -- uvx --from git+https://github.com/BentleySystems/openstaad-mcp openstaad-mcpFor HTTP transport, first start the server in a terminal:
uvx --from git+https://github.com/BentleySystems/openstaad-mcp openstaad-mcp --transport httpLook for the generated token and URL in the terminal output. It should look like this:
WARNING: No --token provided. Auto-generated token: abc123def456ghi789jkl012mno345pq INFO: Starting MCP server 'OpenSTAAD MCP' with transport 'http' (stateless) on http://127.0.0.1:18120/mcpThen add the server in Claude Code with the command:
claude mcp add --transport http openstaad http://127.0.0.1:18120/mcp --header "Authorization: Bearer <your-token>"18120is the default port, but yours may differ if you have multiple instances running or if you changed the default.
Gemini CLI
For stdio transport, use the command:
gemini mcp add openstaad uvx --from git+https://github.com/BentleySystems/openstaad-mcp openstaad-mcpFor HTTP transport, first start the server in a terminal:
uvx --from git+https://github.com/BentleySystems/openstaad-mcp openstaad-mcp --transport httpLook for the generated token and URL in the terminal output. It should look like this:
WARNING: No --token provided. Auto-generated token: abc123def456ghi789jkl012mno345pq INFO: Starting MCP server 'OpenSTAAD MCP' with transport 'http' (stateless) on http://127.0.0.1:18120/mcpThen add the server in Gemini CLI with the command:
gemini mcp add --transport http --header "Authorization: Bearer <your-token>" openstaad http://127.0.0.1:18120/mcp18120is the default port, but yours may differ if you have multiple instances running or if you changed the default.
Transport Modes
The server supports two transport modes:
Mode | When to use |
stdio (default) | The MCP client launches the server process directly. Used by Claude Desktop, Claude Code, VS Code Copilot (stdio config). |
HTTP | The server runs persistently and clients connect over the network. |
CLI Options
Flag | Default | Description |
|
| Transport mode |
|
|
|
| OS default | Path to log file |
|
| [http] TCP port to listen on |
| - | [http] Bearer token for authentication |
Available MCP Tools
Tool | Description |
| Lists available API skills and usage guidance |
| Returns detailed guidance for requested skills |
| Lists active STAAD.Pro instances with model paths and versions |
| Runs validated Python code against the connected STAAD.Pro model |
| Returns connection state, STAAD version, model path, analysis status |
File I/O
The execute_code tool supports optional server-side file I/O for bulk data workflows.
Instead of passing large datasets through the agent's context window, the server reads/writes
CSV and XLSX files directly and injects the data into the sandbox as the input_data variable.
Parameter | Description |
| Path to a |
| Path where the sandbox return value will be written. The return value must be a list-of-lists (CSV) or a |
| Allow overwriting an existing output file (default |
input_data has a stable, extension-specific shape:
CSV: a list of row lists. If a header is detected, it is
input_data[0]and data rows start atinput_data[1:].XLSX: a dict:
{sheet_name: {"columns": list, "rows": list_of_rows}}.
Path containment: File paths must resolve inside a configured allowed boundary before any read/write occurs.
The server supports both client-configured MCP roots and server-configured allowed directories (via --allowed-dirs or user_config.allowed_directories in the manifest).
The server validates paths against these boundaries before any file access.
Limits: Max file size 50 MB, max 100K rows, max 500 columns, max 50 input sheets.
Security Notes
Bearer token authentication. Pass
--token MY_SECRET_TOKENwhen running in HTTP mode and includeAuthorization: Bearer <token>in client requests.DNS rebinding protection. Starlette Middlewares validate
Host,Sec-Fetch-SiteandOriginheaders.Code sandbox. The
execute_codetool validates all Python code via AST analysis before execution. Imports, file access, and dangerous builtins are blocked.
Privacy Policy
Please find the Bentley Systems privacy policy here.
Contributing
See CONTRIBUTING.md for guidelines on setting up your development environment, branch naming, running tests, and submitting pull requests.
Available Tools
5 toolsdiscover_apiDiscover API and skillsARead-onlyIdempotent
Discover available API guidance and skills.
Call this FIRST before using other openstaad-mcp tools.
Then use read_skills with one or more specific skill names to load full guidance.
Inputs: no named parameters. Returns: result
execute_codeExecute Python codeADestructive
Execute Python code in a sandbox against the OpenSTAAD API (don't forget to call discover_api and read_skills for API guidance).
The sandbox provides pre-connected staad (the OpenSTAAD root object) and input_data (if input_data_path is provided) variables (plus json
and math modules). import statements, dir(), getattr(), ... are BLOCKED.
The last expression value or an explicit result = ... assignment is returned as the result.
If output_data_path is provided, the sandbox will write the result to the specified file.
Paths must be on the user LOCAL filesystem and inside MCP roots or configured allowed_dirs.
On Claude Desktop, users can configure allowed directories in the extension settings and Claude can use the filesystem copy_file_to_claude
tool to move files to Claude's filesystem.
Inputs: code (required), instance, overwrite, input_data_path, +1 more. Returns: object
get_statusGet STAAD.Pro instance statusARead-only
Check the connection to a STAAD.Pro instance.
Pass instance (alias from list_instances) to target a
specific instance. Omit it when only one instance is running.
Returns connection state, STAAD version, model path, and the
directories execute_code may currently read/write
(allowed_dirs) — call this again after the user reconfigures
allowed directories, since the MCP server must be relaunched for
that change to take effect and a stale in-context list will
otherwise look correct.
Inputs: instance. Returns: object
list_instancesList running STAAD.Pro instancesARead-only
List all running STAAD.Pro instances.
Returns a list of instances with their alias, process ID, currently
open file path, and STAAD version. Call this before execute_code
when multiple STAAD instances may be running so you can pick the
right one. The alias (e.g. staadPro1) is stable for the
server session even if the model file changes.
If a version is below the minimum supported (25.0.1), a warning
field is included with details about potential data inaccuracies.
Inputs: no named parameters. Returns: result
read_skillsRead OpenSTAAD skillsARead-onlyIdempotent
Read one or more skills by name.
Use discover_api first to list available skills.
Each skill provides domain-specific guidance (e.g. analysis, geometry, loads).
Pass skill names like ["staad-analysis"] or sub-paths like
["staad-steel-design/assets/DESIGN_CODES"] to read reference files
within a skill.
Inputs: skills (required). Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 5 tools
Each tool has a clearly distinct role: discovery/guidance (discover_api, read_skills), instance/connection inspection (get_status, list_instances), and execution (execute_code). Although get_status and list_instances both report version/path details, their core purposes—checking a specific connection vs. enumerating running instances—are clearly differentiated by the descriptions.
All tool names follow a consistent snake_case verb_noun pattern: discover_api, get_status, execute_code, list_instances, read_skills. There are no deviations in style or casing.
Five tools are well-scoped for an MCP server that bridges to OpenSTAAD via code execution. Each tool serves a necessary part of the workflow: guidance discovery, connection/instance management, and code execution.
The surface covers the full lifecycle for this server's purpose: discovering API guidance, reading skills, checking connection/status, listing available STAAD.Pro instances, and executing Python against the OpenSTAAD API. No obvious operational gaps remain for the stated domain.
Maintenance
Related MCP Connectors
Structural FEM for AI agents: frames, plates, buckling, seismic and Eurocode checks, solved locally.
AI Hub for AEC — 50+ 3D formats, clash detection, ACC integration via Autodesk Platform Services.
Read, query, edit and render DWG drawings with AI. No AutoCAD required.
Parametric 3D CAD for AI agents: build print-ready parts, check them, export STL, 3MF or STEP.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables AI agents to interact with Bentley STAAD.Pro models for tasks like load case definition, data extraction, and property setting, running locally with multi-instance support and no cloud dependency.5MIT
- AlicenseAqualityBmaintenanceEnables AI assistants to interact with bridge structural analysis software for modeling, construction stages, and code checking.1562 PyPI1Apache 2.0
- AlicenseNot gradedqualityBmaintenanceEnables AI coding assistants to directly control CSI ETABS for structural and seismic analysis, including model inspection, material/section definition, frame generation, solver execution, and result extraction through natural language.2MIT
- AlicenseNot gradedqualityCmaintenanceEnables AI agents to drive desktop CAD software through natural language with verified execution, failure recovery, and traceable results across modeling, assemblies, drawings, exports, motion, FEA, and related engineering workflows.1MIT