Toolofficial publisher

code_execution (Anthropic)

Tool : A precise action the AI can call · requires anthropic-api

Runs Python and bash in an Anthropic sandbox, with a file editor.

Ideal for data analysis; code execution means it should be treated as a risk even in a sandbox.

Docs

Overall score

6.9 /10

Security: one caveat, see the card

Utility

8.0

Reliability

6.0

Security

One caveat

Ease

5 min · Comfortable

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

Many analysis or calculation tasks go beyond what a language model can reliably handle through reasoning alone: statistics on large files, visualizations, simulations, structured data processing. Anthropic's code_execution tool addresses this need by allowing Claude to run code directly during a conversation, without you having to manage the execution environment yourself.

In practice, the tool is added to an API request and gives Claude access to two sub-tools: Bash command execution and a file editor. Claude writes Python or shell commands, runs them in an isolated server-side container, retrieves the results, and incorporates them into its response. You can send your own files via the Files API (CSV, Excel, JSON, images, text) and retrieve files produced by Claude, such as charts or reports.

This tool is aimed at developers and technical teams building applications via the Anthropic API who need to automate data analysis, complex calculations, or file generation. It also suits those who want to delegate file processing tasks to Claude without exposing their own infrastructure.

However, it is not suited to users looking for a ready-to-use graphical interface: access is through the API only. It is also not appropriate for cases requiring internet access from the container, as it is network-isolated and cannot download additional packages at runtime.

What it offers

·Bash command execution in a secure server-side container, without any action required on your part.
·Built-in file editor allowing Claude to create, read, and modify files, including source code.
·Support for files uploaded via the Files API: CSV, Excel, JSON, XML, images, text files.
·Retrieval of generated files (charts, reports, etc.) via the Files API after execution.
·Container reuse across multiple requests by passing the container ID, with Python state persistence between calls (versions code_execution_20260120 and later).
·Free execution when the tool is used alongside the web_search or web_fetch tools (beyond standard token costs).
·Automatic C2PA signing of image, video, and audio files produced in the sandbox, enabling origin verification.
·Dynamic filtering of web search results directly within the execution environment, at no additional code execution cost.

Our tips

To receive a file generated by Claude, explicitly ask it to copy the file into the $OUTPUT_DIR directory and list that directory in the same command: this confirms the file was captured before you download it.
Reuse the same container across multiple requests by passing the container ID returned in the response: this preserves files and, with recent versions, Python variable state.
If you need Claude to run code for a simple task it might handle through reasoning, ask explicitly (for example: 'run code to verify this result') to force use of the tool.
Prefer version code_execution_20260521 or code_execution_20260120 if you need Python state persistence or programmatic tool calling from within the sandbox; version code_execution_20250825 is sufficient for basic Bash and file operations.
Combine the tool with web_search or web_fetch to benefit from code execution at no additional cost beyond tokens, while also taking advantage of dynamic result filtering.

Watch out

The container has no internet access and cannot download additional packages at runtime: only pre-installed libraries are available, which may block certain use cases.
Files written outside the $OUTPUT_DIR directory are not returned automatically: if you forget to ask Claude to copy them there, you will need to send another request on the same container to retrieve them.

In practice

Price: Freemium
Commercial use: Commercial use: yes
Available on: api
Host: anthropic-api
FR interface: No

Security and data

Rating: Security: one caveat, see the card
Capabilities: readswritesexecutesnetwork
Your data: Data with the vendor
Maintenance: Active

Links and repository checked on 25 October 2026 · added on 1 May 2025

How to install code_execution (Anthropic)

1. Obtain an API key at platform.claude.com and install the Anthropic SDK for your language (Python, TypeScript, etc.). 2. Add the tool to your API request with the type field set to the desired version (for example: code_execution_20250825) and name fixed to code_execution. 3. To send your own files, upload them first via the Files API and reference their ID in a container_upload block in your message. 4. Retrieve files produced by Claude by extracting the file_id entries from the result blocks, then download them via the Files API. 5. To reuse the same container across multiple requests, pass the container ID returned in the response via the container parameter of the next request. 6. Precaution: the container is isolated from your machine and the network, but treat any executed code as a sensitive operation and review generated files before using them in production.

tools=[{"type":"code_execution_20260120","name":"code_execution"}]

Official documentation ↗

In the same category

Whole category Feed (memory, context) →