Skip to main content

Usage

Learn how to use MCP Jupyter effectively with your AI assistant.

Basic Usage​

Creating a New Notebook​

Ask your AI assistant to create a notebook:

"Create a new notebook called data_analysis.ipynb"

The AI will:

  1. Create the notebook file
  2. Start a kernel
  3. Be ready for your commands

Working with Existing Notebooks​

"Open the notebook experiments/model_training.ipynb"

Your AI assistant will connect to the existing notebook and preserve all current state.

Example Workflows​

Data Analysis Workflow​

Here's a typical data analysis session:

# You: Load the sales data
# AI: *loads data and shows preview*

# You: "I see there are missing values. Handle them appropriately"
# AI: *analyzes data types and fills missing values*

# You: Manually explore specific columns
df['revenue'].describe()

# You: "Create visualizations for the quarterly trends"
# AI: *generates comprehensive visualizations*

Machine Learning Workflow​

# You: "Load the iris dataset and prepare it for classification"
# AI loads data, does train-test split

# You manually inspect the data
X_train.shape, y_train.value_counts()

# You: "Try different classifiers and compare their performance"
# AI implements multiple models with cross-validation

# If a package is missing, AI will see the error and install it
# automatically, then retry the operation

Package Management​

The AI assistant handles package installation seamlessly:

# You: "Create a word cloud from this text"
# AI attempts to import wordcloud
# Sees ImportError
# Installs the package: !pip install wordcloud
# Retries and creates the visualization

Advanced Features​

State Preservation​

All variables remain available throughout your session:

# Cell 1 (executed by you)
data = load_large_dataset()
model = train_complex_model(data)

# Cell 2 (AI continues with your objects)
# AI can access 'data' and 'model' directly
# "Evaluate the model and show feature importance"

Error Handling​

The AI can see and respond to errors:

# You write code with an error
result = data.groupby('category').mean()  # Error: 'data' not defined

# AI sees the error and can:
# - Suggest loading the data first
# - Check available variables
# - Provide the correct code

Collaborative Exploration​

Switch seamlessly between manual and AI work:

# You: Start exploring
df.head()
df.info()

# You: "Continue exploring this dataset and find interesting patterns"
# AI: Performs statistical analysis, creates visualizations

# You: Notice something interesting in the AI's output
subset = df[df['category'] == 'electronics']
subset['profit_margin'].hist()

# You: "Investigate why electronics have this distribution"
# AI: Continues analysis focusing on your discovery

Best Practices​

1. Clear Instructions​

Be specific about what you want:

  • ❌ "Analyze the data"
  • ✅ "Perform exploratory data analysis focusing on customer segments and seasonal patterns"

2. Iterative Refinement​

Work iteratively with the AI:

1. "Load and preview the customer data"
2. Review the output
3. "Focus on customers from the last quarter"
4. "Now segment them by purchase frequency"

3. State Management​

  • Keep important variables in the global namespace
  • Use descriptive variable names
  • Periodically check available variables with dir() or locals()

4. Error Recovery​

When errors occur:

  • Let the AI see and handle the error
  • Provide context if needed
  • The AI will install packages or fix issues automatically

Demo Example​

MCP Jupyter Demo

View the generated notebook →

Tips and Tricks​

  1. Use Markdown cells: Ask the AI to document its analysis
  2. Save checkpoints: Periodically save important state
  3. Combine approaches: Use AI for boilerplate, manually tune details
  4. Leverage errors: Let errors guide package installation
  5. Incremental development: Build complex analyses step by step

Next Steps​