Anthropic is announcing four new capabilities on the Anthropic API that enable developers to build more powerful AI agents: the code execution tool, the MCP connector, the Files API, and the ability to cache prompts for up to one hour.

Building Better AI Agents

With these new beta features, combined with Claude Opus 4 and Sonnet 4, developers can now create AI agents that are smarter, more efficient, and capable of doing more on their own. These agents can now run code to analyze data, connect to other systems, handle files across sessions, and remember what’s going on for up to 60 minutes, all without needing to build complex infrastructure from scratch.

Example in Action: Imagine a project management AI that connects to Asana using the MCP connector, uploads reports through the Files API, analyzes project risks using the code execution tool, and remembers all this work through extended prompt caching. It’s efficient, powerful, and cost-effective.

These new features build on existing tools like web search and citations, giving developers everything they need to create truly helpful AI agents. Let’s break down what each feature does.

Code Execution Tool

This tool allows Claude to run Python code in a safe, sandboxed environment. Instead of just suggesting code, Claude can now,

• Analyze data
• Create charts
• Clean datasets
• Discover patterns

All of this happens within the API itself. Claude becomes more than just a coding assistant; it can now act like a full-on data analyst.

Key uses include.

• Financial Modeling: Projecting investments and analyzing portfolios.
• Scientific Computing: Running simulations and analyzing data.
• Business Intelligence: Creating reports and dashboards.
• Document Processing: Automating and formatting data from documents.
• Statistical Analysis: Running regressions and predictive models.

Organizations get 50 free hours per day, and after that, it’s $0.05/hour per container.

MCP Connector

This makes it easier than ever to connect Claude to external systems using the Model Context Protocol (MCP). Before, developers had to build their own code to make these connections. Now, the Anthropic API handles everything automatically.

Here’s how it works.

• Claude connects to the listed MCP servers
• Finds and uses available tools
• Decides what to do and executes the task
• Handles login and errors
• Returns a complete response with integrated data

You can even connect to popular tools like Zapier and Asana instantly. No extra code is required.

Files API

Managing documents just got easier. With the new Files API, developers can upload files once and reuse them across multiple conversations or sessions.

Why is it useful?

• No need to re-upload big files each time
• Great for projects involving lots of documents or datasets
• Works directly with the Code Execution Tool for seamless analysis and output (like charts and graphs)

Now, Claude can access and analyze uploaded files as needed, saving time and effort.

Extended Prompt Caching

Prompt caching now lasts up to 1 hour, instead of the usual 5 minutes. This major improvement helps developers.

• Reduce costs by up to 90%
• Cut latency by up to 85%
• Maintain context over longer workflows

Perfect for agents that work through complex processes, analyze long documents, or integrate across systems over time. This makes long-running AI projects far more practical and scalable.

Get Started Today

All four features are now available in public beta on the Anthropic API. Whether you're building with Claude, Claude Code, or another platform, it’s a great time to start.

Visit the documentation or check out the developer keynote to see these tools in action. Anthropic is also aiming for higher AI Safety Levels like ASL-3—another step toward building trustworthy and capable AI.