Add Matplotlib Agent Toolkit

[SyedAbdullahh]

Hi team,

This PR introduces a new Matplotlib Agent Toolkit to langchain_experimental/agents/agent_toolkits/matplotlib. It’s something I mentioned earlier in Issue #63 . The goal is to let LLMs generate Matplotlib visualizations directly from DataFrames through natural language.

What’s included

• base.py: main logic for creating a Matplotlib-based agent
• prompt.py: templates and example prompts
• tests/test_matplotlib_toolkit.py: 7 unit tests (all passing)
• README.md: usage guide, parameters, and security notes

What it does

The agent works with any LLM that supports tool-calling (like ChatOpenAI) and can handle one or multiple Pandas DataFrames.
It builds a custom prompt, uses PythonAstREPLTool for safe inline code execution, and returns an AgentExecutor that can turn natural language queries into plots.

Testing

You can run:

pytest  libs/experimental/tests/unit_tests/agents/agent_toolkits/matplotlib/test_matplotlib_toolkit.py -v

All tests pass locally — they cover agent creation, validation, error handling, and multiple DataFrame support.

Security

This agent uses a Python REPL, so allow_dangerous_code=True must be explicitly set by the user.
I’ve added clear warnings and notes in the README about sandboxing and safe environments.

@efriis, @ccurme

[SyedAbdullahh]

@ccurme can you please review my pull request?