AI Agent pro

This page of the data inspector lets you connect an external AI coding agent (e.g. Claude Code, GitHub Copilot CLI, OpenAI Codex, and others) to your running OVITO Pro session. The agent runs right inside the OVITO main window and has live access to the scene you are working on – it can inspect the data pipelines you have built, run Python code against them, and see the same viewports and panels you see.

This turns OVITO Pro into an interactive AI-assisted workspace: instead of switching back and forth between OVITO and a separate chat window, you can ask the agent to write and run Python scripting code for you, have it help track down why a modifier produces an unexpected result, or simply describe an analysis or visualization task in plain language and let the agent turn it into a working data pipeline. Because the agent operates on the exact same session you are looking at, you can take over at any point, and it can pick up again from whatever state you left the scene in.

How it works

The integration is built on the Model Context Protocol (MCP), an open standard that lets AI coding agents call out to external tools. It has three parts:

The coding agent is a third-party command-line program (Claude Code, Copilot CLI, Codex, etc.) that you install and configure yourself, exactly as you would for any other project. This applet simply launches that program for you inside an embedded terminal, docked into OVITO’s main window for easy access.
The ovito-mcp MCP server is a small Python package, developed by the OVITO team, that you add to your coding agent’s list of available MCP tool providers (consult your agent’s documentation for how to configure an MCP server). It is the bridge between the agent and OVITO: the agent calls tools such as run this Python code, take a screenshot of the window, or list the pipelines in the scene, and the server translates these calls into commands sent to OVITO.
OVITO Pro’s embedded Python interpreter is what actually executes those commands, in the same process and on the same scene as the OVITO window you see on screen. When you click Start, OVITO starts a private Jupyter kernel inside itself and passes its connection details to the newly launched agent process as part of its initial prompt, so the agent knows how to reach it through the ovito-mcp server.

Put together, this gives the agent two complementary ways of “seeing” what you are doing: visually, by taking a screenshot of the OVITO window on request, and structurally, by directly inspecting the live data collections, pipelines, and modifiers in your scene. Because all of this happens against the same running session, you and the agent are effectively looking over each other’s shoulder and can solve problems together interactively, rather than the agent working blind on a description of the problem.

Note

The ovito-mcp package is a separate download, distributed via PyPI. See its accompanying documentation for installation and configuration instructions specific to your coding agent.

Launching a coding agent

The panel offers the following controls:

Coding agent

A drop-down list of coding agent CLIs that OVITO knows how to launch out of the box, each with the command line arguments it requires to receive the initial connection prompt. Pick one of these presets, or select Custom… to connect an agent that isn’t in this list.

Choosing Custom… reveals a Command line field where you enter the full command used to launch the agent, including its executable name (or full path, if it is not on your system’s PATH) and any command line arguments it needs. The command line must include the placeholder <PROMPT> somewhere in it; OVITO replaces this placeholder with the actual connection prompt when it starts the process. The field is initially filled in with the equivalent command line of whichever preset agent you last used, as a convenient starting point, and remembers whatever you subsequently type into it.

Working directory

The directory the agent process is started in, shown by Change… to pick a different one. This is typically relevant if the agent inspects or modifies files in your project folder in addition to talking to OVITO.

Start

Launches the selected coding agent. The panel switches to an embedded terminal showing the agent’s own text interface, which you can interact with directly, just like you would in a regular terminal window. A small close button in the top-right corner of the terminal ends the session and returns to the launcher form; the session also ends automatically once the agent process exits on its own.

Note

Only one coding agent session can be active at a time, even if you have multiple OVITO windows open, since the embedded Python interpreter is shared by the entire OVITO Pro process.

Security considerations

Connecting a coding agent to OVITO gives it the ability to execute arbitrary Python code with the same privileges as your OVITO Pro session, including full access to the file system, network, and any data loaded in the current scene. Keep the following in mind:

Only connect coding agents and MCP servers that you trust. The ovito-mcp server itself does not add any access restrictions on top of what the agent can already do on your machine.
As with any AI agent that can call tools autonomously, review consequential actions it proposes (e.g. deleting or overwriting files, running exports, making network requests) before letting it proceed, and be mindful of prompt injection risks if the agent processes untrusted external data.
If your coding agent uses a cloud-hosted LLM, be aware that code, error messages, and screenshots exchanged with the agent may be transmitted to that provider. Take this into account if you are working with sensitive or proprietary data.

Give us feedback

This AI agent integration is a new addition to OVITO Pro, and we would love to hear from early adopters about your experience with it: what worked well, what didn’t, and what would make it more useful for your workflows. Please send your feedback, bug reports, or suggestions to support@ovito.org.