Why does a chart come back as a file rather than appearing inline?

Because the code runs remotely, plt.show() has no display to render into. Scripts write the image to /home/user and Get File returns it instead. The same rule applies to any output worth keeping: write it to a file rather than relying on a rendered cell.

Where do files have to sit inside the sandbox?

In /home/user. Upload File puts an attachment there, Execute Code and Run Terminal Command read and write there, and Get File reads from there, so a path outside that folder is the usual reason a script cannot find its input.

What happens when the code errors out?

Execute Code returns stderr and the error alongside stdout, so a failure comes back with its traceback rather than a blank result. That lets the script be corrected and rerun in the same sandbox, and anything still unresolved goes to a human agent with the full conversation history attached.

Is this a safe place to run something against customer data?

The sandbox is a separate cloud environment created on demand by Create Sandbox, not your own infrastructure, so nothing executes on your servers. Anything sensitive only gets there because Upload File deliberately put it there, so treat what you upload with the same care as data sent to any third-party processor.

Codeinterpreter MCP integration

Spins up a Python sandbox, uploads files into it, runs code or shell commands, and returns the file that comes out.

Connect Codeinterpreter See all integrations

5actions available

Three actions you can hand over today

Every action runs live through MCP. Nothing to build, nothing to maintain.

Execute code
Lyro executes Python code in a sandbox and returns the results, stdout, stderr, and any errors. Support agents can run calculations or tests without leaving the conversation, solving problems on the fly.
Run terminal command
Lyro runs a terminal command in the sandbox environment and returns the output and exit code. Agents can perform system operations, check file states, or run scripts without leaving the support context.
Create sandbox
Lyro creates a new sandbox environment for executing Python code in a Jupyter notebook cell. Each conversation gets an isolated workspace, ensuring customer data stays separate and sessions don't interfere.

See all 5 actions

How businesses use Codeinterpreter + Lyro

Each card is one request a support team gets, and the Codeinterpreter actions Lyro runs to close it.

Run real Python on the numbers in a question
Lyro creates a sandbox, executes the code in a Jupyter cell and reads back the result, stdout, stderr and any error, dropping to a terminal command when a package install or a file listing is needed first.
Create SandboxExecute CodeRun Terminal Command
Take a file in and hand a file back
Lyro uploads the attachment into /home/user, works on it inside the sandbox, and returns the CSV or chart image the script wrote there, which is how visual output travels when the code runs remotely.
Upload FileGet File

How it works

Get started in 3 steps

Connect once, then just ask. There is no workflow builder to learn and nothing to maintain — Lyro reads the Codeinterpreter actions it has and picks the ones a request needs.

01
Connect Codeinterpreter
Authorize the Codeinterpreter account your team already uses — one consent screen, no API keys, no mapping tables. Lyro can only do what you granted that account, and you can disconnect it at any time.
02
Tell your agent what you need
Describe the job the way you would hand it to a teammate. Lyro maps it to the Codeinterpreter actions that close it and chains as many as the request needs.
03
Watch it work
The agent runs the actions inside the conversation the customer is already in, so nobody copies data between tabs and your team can take over at any point.
Get started free

Codeinterpreter · Lyro

Everything else about Codeinterpreter

Setup, permissions, and the limits of what Lyro can do inside Codeinterpreter.

Because the code runs remotely, plt.show() has no display to render into. Scripts write the image to /home/user and Get File returns it instead. The same rule applies to any output worth keeping: write it to a file rather than relying on a rendered cell.

Every action available in Codeinterpreter

All 5 actions your agent can call on Codeinterpreter, straight from the live MCP connection.

Create sandbox
Create a sandbox to execute python code in a Jupyter notebook cell.
Execute code
Execute python code in a sandbox and return any result, stdout, stderr, and error.
Get file
Get a file from the sandbox and returns the file.
Run terminal command
Run a command in the terminal and returns the stdout, stderr, and error code.
Upload file
Upload a file to the sandbox environment.