Tools
A tool is a separate operation that the agent calls through Engee MIND with formal arguments. The user usually formulates the task in text; the agent selects a tool, passes the parameters, and analyzes the response.
There are currently 17 tools available. The LLM client can add a server prefix to the technical names.
Tool Reference
The table shows the subject arguments. For the first check, send the PAT to the chat with the agent; the issue is described in external API documentation Engee.
| The tool | Task Arguments | Result and purpose |
|---|---|---|
|
No |
The state of the environment; without PAT prompts for authorization and built—in skills. Called first |
|
No |
Launches the environment and waits for it to be ready; returns information about the found user skills |
|
No |
Stops the work environment, frees up computing resources |
|
|
Executes the Julia code in the session; returns a string representation of the last expression |
|
|
Creates a long-running independent background task and returns |
|
|
Returns the state |
|
No |
Shows the background tasks of the current user session |
|
Optional |
List of files and directories in the session |
|
|
Uploads the contents of the file to the session using the absolute path |
|
|
Returns the name, size, and contents of a file from the session for saving or displaying by the client |
|
|
Model status, logs, blocks, port connections, and the model script |
|
|
Creates a PNG of the canvas in the session and returns its path; the model should be open |
|
No |
Block library categories and their contents |
|
|
Searches for blocks by name, category, or keyword; returns up to 30 matches per call |
|
|
Exact parameters, acceptable values, ports, and a sign of direct signal transmission |
|
No |
A list of available skills and their sources; takes into account user skills in the session |
|
|
The text of the skill or its help file |
How the tools are combined
Not every task needs all the steps. To view the help, you can download a common skill without running a custom environment.
Examples of arguments
Block Search:
{
"query": "integrator",
"offset": 0
}
Uploading a basic skill:
{
"skill_path": "engee_model"
}
Diagnostics of an already open model in an authorized environment:
{
"model_name": "MyModel",
"server": "https://engee.com",
"token": "<PAT_ENGEE>"
}
<PAT_ENGEE> — a placeholder, not a valid token. These fragments describe the arguments for invoking the tool, not the configuration of the MCP client.
Features of execution and files
-
engee_stopstops the entire work environment. Unsaved models and variables in memory are lost; files saved in/user/, remain available. Challengeengee.stop()inside, Julia stops the simulation, not the session. -
job_createit is executed in a separate context. Variables fromeval_codethey are not automatically transferred there. A long-running task should load the inputs itself, perform the calculation, and save the results. -
file_downloaddownloads a file, not a directory. The Base64 content must be decoded by the client into a file; a long Base64 string in a chat is not a convenient output. -
engee_screenshotcreates a snapshot of the circuit; to transfer the image to the user, then you needfile_downloadon the way back. -
eval_codemodifies the model and files depending on the transmitted code. The permission for this tool actually allows the agent to perform calculations in the user environment; keep this in mind when configuring client rights.
The current list of a particular server takes precedence over the fixed number in this edition.