OriginLab数据分析与可视化MCP服务器

OriginLab MCP Server

Use AI as your OriginLab assistant

Connect OriginLab data analysis and visualization to Antigravity, Claude, Cursor, and other AI clients through the Model Context Protocol.

Quick Start · Features · Examples · Client Configuration

简体中文 · English

What Is OriginLab MCP Server?

OriginLab MCP Server is a bridge between AI clients and OriginLab. It lets you import data, create plots, customize figures, run analysis, and export results through natural-language requests instead of manually operating the Origin UI.

User: Import experiment.csv from my desktop into Origin, use the first column as X
      and the second column as Y, create a scatter plot, run a Gaussian fit,
      then export the graph as PNG.

AI: Done.
    Imported experiment.csv -> Sheet1 (200 rows x 5 columns)
    Created scatter plot -> Graph1
    Gaussian fit completed -> xc=2.35, w=0.82, A=156.3, R²=0.9987
    Exported -> C:\Users\Desktop\Graph1.png

How It Works

AI Client (Antigravity / Claude / Cursor)
       ↓ MCP over stdio
OriginLab MCP Server (Python)
       ↓ originpro + COM
OriginLab

Requirements

Quick Start

For AI assistants or one-command local setup, run this from the repository root:

powershell -ExecutionPolicy Bypass -File .\scripts\install-and-open.ps1

The script installs uv if needed, runs uv sync, starts the local status panel, and opens http://127.0.0.1:8765/ automatically. From that page you can test Origin and write MCP client configs.

Choose one installation method.

irm https://astral.sh/uv/install.ps1 | iex

cd C:\path\to\originlab-mcp
uv sync

uv run originlab-mcp

cd C:\path\to\originlab-mcp
python -m venv .venv
.venv\Scripts\activate

pip install -e .

originlab-mcp

After startup, the server waits for MCP client requests over stdio. The first tool call automatically connects to the local OriginLab installation.

Optional: Local Status Panel

To check whether the MCP server can start, whether Origin can be reached, and whether common client config files exist, start the local UI:

uv run originlab-mcp-ui

Then open http://127.0.0.1:8765/. This page can start/stop a debug MCP server subprocess, test the Origin connection, read the current Origin session (worksheets and graphs), and write the originlab MCP configuration for Antigravity / Gemini, Cursor, Codex, Trae, and Claude Desktop. For normal use, the AI client should still start the server automatically from its configuration.

When updating an existing config file, the UI creates a .bak-timestamp backup first. To prevent the browser from opening automatically:

$env:ORIGINLAB_MCP_UI_NO_BROWSER = "1"
uv run originlab-mcp-ui

Features

The server provides 67 tools covering the OriginLab data workflow.

Data Management (14 Tools)

Plotting (11 Tools)

Graph Customization (25 Tools)

Analysis (3 Tools)

linear_fit · nonlinear_fit · list_fit_functions

Common Origin fit functions such as Gauss, Lorentz, ExpDec1, and Boltzmann are supported. You can provide initial parameters, fix parameters, and fit with error bars.

Export and Project Management (6 Tools)

export_graph · export_all_graphs · export_worksheet_to_csv · save_project · open_project · new_project

System Management (6 Tools)

get_origin_info · originlab_doctor · read_origin_session · release_origin · reconnect_origin · close_origin

originlab_doctor is a read-only environment checklist (platform, originpro, autosave / dispatch / allowed-roots policy). Pass ping_origin=true to attempt a live Origin connect.

read_origin_session is a read-only snapshot of the current project: workbooks/worksheets, graphs, matrices, notes, active objects, and project path. Pass include_preview=true to include a truncated preview of the active worksheet.

MCP Resources (Session Reading)

In addition to tools, clients can inspect the Origin session through MCP resources/read without changing the project:

MCP Prompts (Workflow Templates)

Clients that support prompts/list can start common Origin workflows from templates:

Advanced (2 Tools)

execute_labtalk · get_labtalk_variable

execute_labtalk is an escape hatch for operations not covered by standard tools. get_labtalk_variable safely reads LabTalk variable values. Destructive LabTalk (delete / doc -n / win -c, etc.) requires confirm=true.

Safety: Preflight Autosave

Before destructive tools (new_project, open_project, clear_worksheet, delete_columns, remove_plot_from_graph, remove_graph_label, close_origin, and confirmed destructive LabTalk), the server attempts an in-place project save when a project path is known:

Save explicitly with save_project(file_path=...). Implicit saves and autosave now require a unique, full project path from the live session; cached destinations are never reused after a manual project switch or reconnect. Missing or ambiguous metadata skips autosave and blocks destructive operations in required mode. Some Origin versions may not expose the currently supported path metadata; use an explicit save path and record the result in the Windows acceptance checklist. In-place saving overwrites the file and is not a versioned backup. Preflight save and the following mutation are not yet an atomic transaction.

Safety: Soft Dispatch Timeout

Each COM-backed tool call has a soft wall-clock budget (default 90s). If Origin does not return in time — often because a modal dialog is open — the server raises a structured timeout error without killing Origin. The underlying COM call may still be running and will hold the lock until Origin unblocks.

execute_labtalk(..., timeout=120) overrides the budget for that call only (0 disables).

Safety: Allowed Roots (optional)

When set, all import / export / save / open file paths must resolve under one of the allowed directories (path traversal via .. is rejected). Unset keeps unrestricted access for typical desktop Origin workflows.

get_origin_info reports the active allowlist as allowed_roots.

Examples

apply_publication_style supports layer_index, axis-title font size, tick-label font size, legend font size, major tick length, minor tick count, line width, symbol size, and other common publication-figure settings.

Typical workflow:

Import data -> inspect structure -> set column designations -> create plot -> customize graph -> run analysis -> export results

Client Configuration

[!NOTE] You do not need to start the MCP server manually. Once configured, the AI client starts the server process when needed and communicates with it over stdin/stdout.

Replace paths with your actual project path.

You can also run uv run originlab-mcp-ui, choose a client in the local status panel, and let it write the config automatically. Manual examples are shown below.

Antigravity (Gemini) - Recommended

Create .gemini/settings.json in the project root.

Using uv:

{
  "mcpServers": {
    "originlab": {
      "command": "uv",
      "args": ["--directory", "C:\\path\\to\\originlab-mcp", "run", "originlab-mcp"]
    }
  }
}

Using pip:

{
  "mcpServers": {
    "originlab": {
      "command": "C:\\path\\to\\originlab-mcp\\.venv\\Scripts\\originlab-mcp.exe"
    }
  }
}

[mcp_servers.originlab]
command = "uv"
args = ["--directory", 'C:\path\to\originlab-mcp', "run", "originlab-mcp"]

Testing

# uv
uv run python -m pytest tests/ -v

# pip, after activating the virtual environment
pytest tests/ -v

The basic test suite does not require OriginLab to be installed. On Linux/macOS, originpro is skipped automatically because it is marked Windows-only in pyproject.toml. GitHub Actions runs ruff, mypy, and pytest (with coverage XML) on Python 3.10–3.12.

Project Structure

originlab-mcp/
├── pyproject.toml                # Project configuration and dependencies
├── CHANGELOG.md                  # Release notes
├── .github/workflows/ci.yml      # Lint / typecheck / unit tests
├── scripts/
│   └── install-and-open.ps1      # One-command setup and UI launcher
├── src/originlab_mcp/
│   ├── server.py                 # MCP server entry point and dependency injection
│   ├── ui.py                     # Local status panel
│   ├── origin_manager.py         # Thread-safe Origin COM connection manager
│   ├── session.py                # Read-only Origin session snapshot
│   ├── resources.py              # MCP Resources (session reading)
│   ├── prompts.py                # MCP Prompts (workflow templates)
│   ├── exceptions.py             # Custom exceptions
│   ├── types.py                  # Protocol type definitions
│   ├── tools/
│   │   ├── data.py               # Data import and worksheet management (14)
│   │   ├── plot.py               # Plot creation and graph management (11)
│   │   ├── customize.py          # Graph customization (25)
│   │   ├── analysis.py           # Linear and nonlinear fitting (3)
│   │   ├── export.py             # Export and project management (6)
│   │   ├── system.py             # System and connection management (6)
│   │   └── advanced.py           # LabTalk escape hatch (2)
│   └── utils/
│       ├── annotations.py        # MCP ToolAnnotations presets
│       ├── autosave.py           # Preflight autosave policy
│       ├── dispatch.py           # Soft COM dispatch timeout
│       ├── doctor.py             # originlab_doctor diagnostics
│       ├── paths.py              # Allowed-roots path sandbox
│       ├── labtalk_safe.py        # LabTalk confirm-gate scanner
│       ├── constants.py          # Enums, defaults, and fit-function metadata
│       ├── helpers.py            # Graph/sheet resolution and error handling helpers
│       └── validators.py         # Input validation and standard response builders
└── tests/
    ├── conftest.py               # Shared fixtures (manager / fake Origin)
    ├── fakes/                    # In-memory OriginPro + DummyMCP doubles
    ├── test_annotations.py       # ToolAnnotations coverage tests
    ├── test_autosave.py          # Autosave policy / preflight tests
    ├── test_dispatch.py          # Soft dispatch timeout tests
    ├── test_paths.py             # Allowed-roots sandbox tests
    ├── test_prompts.py           # MCP prompt workflow templates
    ├── test_doctor.py            # originlab_doctor diagnostics
    ├── test_origin_contract.py   # Multi-tool COM contract workflows
    ├── test_tool_guidance.py     # Description / next_suggestions contracts
    ├── test_helpers.py           # Helper tests
    ├── test_phase3.py            # LabTalk safety and resolve-pattern tests
    ├── test_session.py           # Session reading and MCP resource tests
    ├── test_tools.py             # Tool registration and integration tests
    └── test_ui.py                # Local status panel config tests

FAQ

License

MIT © 2025 garethbeaumo

For real stdio and Windows + Origin checks, see the acceptance guide and evidence-driven follow-up.

Queued calls whose budget expires before acquiring the COM lock never execute; already-started calls may still finish after a soft timeout.