Quick Start

To get started, run the following command:

bash scripts/configure_mcp.sh

This script will guide you through the process of setting up your environment variables and configuring the MCP connections for both Claude and Gemini.

Palantir Foundry AI-Assisted Development Template

A comprehensive template for developing Palantir Foundry applications with integrated AI code assistants (Claude Code and Google Gemini CLI).

Prerequisites

• Node.js (v18 or higher)
• npm or yarn
• Access to Palantir Foundry with:
  • Platform administrator has enabled Palantir MCP
  • Generated user token
  • Your enrollment URL (e.g., stack123.palantirfoundry.com)
  • Compass project RID for default resource creation

Initial Setup

1. Clone and Configure

# Clone this template
git clone <your-repo-url> my-palantir-project
cd my-palantir-project

# Set up environment variables
export FOUNDRY_TOKEN="your-palantir-user-token"

2. Configure npm Registry

Configure npm to authenticate with Palantir's artifact registry:

npm config set //$(echo "https://your-enrollment.palantirfoundry.com/artifacts/api/repositories/ri.artifacts.repository.discovered.foundry-mcp/contents/release/npm/" | sed 's/https:\/\///')/:_auth $FOUNDRY_TOKEN

Replace your-enrollment with your actual enrollment URL.

3. Update Configuration Files

For Claude Code

Edit .mcp.json and replace the placeholders:

{
  "mcpServers": {
    "Palantir": {
      "command": "npx",
      "args": [
        "-y",
        "@palantir/mcp@latest",
        "--foundry-api-url",
        "https://your-enrollment.palantirfoundry.com",
        "--default-foundry-folder",
        "ri.compass.folder.your-project-rid"
      ],
      "env": {
        "FOUNDRY_TOKEN": "your-token-here",
        "NPM_CONFIG_REGISTRY": "https://your-enrollment.palantirfoundry.com/artifacts/api/repositories/ri.artifacts.repository.discovered.foundry-mcp/contents/release/npm/"
      },
      "disabled": false
    }
  }
}

For Gemini CLI

Create or update ~/.gemini/settings.json:

mkdir -p ~/.gemini
cat > ~/.gemini/settings.json << 'EOF'
{
  "mcpServers": {
    "Palantir": {
      "command": "npx",
      "args": [
        "-y",
        "@palantir/mcp@latest",
        "--foundry-api-url",
        "https://your-enrollment.palantirfoundry.com",
        "--default-foundry-folder",
        "ri.compass.folder.your-project-rid"
      ],
      "env": {
        "FOUNDRY_TOKEN": "your-token-here",
        "NPM_CONFIG_REGISTRY": "https://your-enrollment.palantirfoundry.com/artifacts/api/repositories/ri.artifacts.repository.discovered.foundry-mcp/contents/release/npm/"
      }
    }
  }
}
EOF

Remember to replace:

• your-enrollment with your Palantir enrollment URL
• your-token-here with your generated user token
• ri.compass.folder.your-project-rid with your Compass project RID

4. Verify Setup

Claude Code:

# Start Claude in the project
claude

# List MCP servers
claude mcp list

# Test Palantir connection
> Can you list the datasets in my Foundry folder?

Gemini CLI:

# Check MCP configuration
gemini /mcp

# Test in agent mode
gemini --agent
> Show me the datasets in my default Foundry folder

Project Structure

.
├── .claude/                    # Claude Code configuration
│   ├── commands/              # Custom Claude commands
│   ├── settings.json          # Base permissions
│   └── settings.local.json    # Local permissions
├── .mcp.json                  # MCP server configuration
├── ai_docs/                   # AI assistant documentation
│   ├── palantir-patterns.md   # Common Foundry patterns
│   └── foundry-best-practices.md # Best practices guide
├── specs/                     # Project specifications
│   ├── code-standards.md      # Code standards
│   ├── project-management.md  # Project management
│   └── project_plan.md        # Master project plan
├── foundry/                   # Foundry-specific code
│   ├── transforms/           # Data transformation code
│   ├── functions/           # Foundry Functions
│   └── pipelines/           # Pipeline definitions
├── scripts/                   # Utility scripts
└── CLAUDE.md                  # Claude-specific instructions

Using with Palantir Foundry

Available MCP Tools

Once connected, AI assistants have access to:

• Dataset Operations: Read, write, query, and transform datasets
• Code Repository Management: Create and manage Foundry code repositories
• Pipeline Execution: Trigger builds and monitor pipeline status
• Ontology Access: Query and manipulate ontology objects
• Function Execution: Run Foundry Functions
• File Management: Upload and manage files in Foundry

Example Commands

Dataset Operations:

# With Claude Code
> Show me the schema of the customer_transactions dataset
> Create a transform that aggregates daily sales by region
> Write a Python transform to clean the raw_inventory data

# With Gemini CLI
> List all datasets modified in the last week
> Generate a PySpark transform to join customer and order datasets

Pipeline Management:

# With Claude Code
> Check the status of the daily_aggregation pipeline
> Create a new pipeline that runs the data quality checks

# With Gemini CLI
> Show me failed builds from the last 24 hours
> Set up a pipeline to run every 6 hours

Ontology Operations:

# With Claude Code
> Query all Customer objects where revenue > 1000000
> Create a new ontology type for Product with appropriate properties

# With Gemini CLI
> Show the relationships between Customer and Order objects
> Update the Customer object type to include a new property

Development Workflow

1. Starting a New Feature

# Load project context
claude
> /prime

# Work on a JIRA ticket
> /work_on_ticket_engineer PROJ-123

2. Data Transformation Development

# Create a new transform
> Create a new Python transform in foundry/transforms/ that processes daily_sales data

# The AI will:
# - Create the transform file
# - Set up proper imports
# - Implement the transformation logic
# - Add error handling
# - Create unit tests

3. Pipeline Development

# Create a pipeline
> Set up a pipeline that orchestrates the customer data processing workflow

# The AI will:
# - Define pipeline stages
# - Set up dependencies
# - Configure scheduling
# - Add monitoring and alerts

Best Practices

Security

1. Never commit tokens: Use environment variables for sensitive data
2. Token rotation: Regularly rotate your Foundry user tokens
3. Access control: Follow principle of least privilege
4. Code review: Always review AI-generated code before deployment

Development

1. Use version control: Commit changes frequently with meaningful messages
2. Test thoroughly: Ensure transforms work with sample data before full runs
3. Monitor performance: Check dataset size and processing times
4. Document changes: Keep README and specs updated

AI Assistant Usage

1. Be specific: Provide clear requirements and constraints
2. Verify outputs: Always review generated code and queries
3. Iterative development: Build complex transforms incrementally
4. Use examples: Provide sample data or expected outputs when possible

Customization

Adding Project Context

Edit CLAUDE.md to include:

• Specific dataset schemas and relationships
• Common transformation patterns used in your project
• Ontology object definitions
• Team conventions and standards

Extending AI Capabilities

Add documentation in ai_docs/:

• palantir-patterns.md - Common Foundry development patterns
• dataset-schemas.md - Dataset schemas and relationships
• transform-library.md - Reusable transformation code
• ontology-guide.md - Ontology structure and usage

Troubleshooting

MCP Connection Issues

1. Verify token is set correctly:

   echo $FOUNDRY_TOKEN

2. Check npm configuration:

   npm config get registry

3. Test MCP directly:

   npx -y @palantir/mcp@latest --foundry-api-url https://your-enrollment.palantirfoundry.com --help

Common Issues

• "Unauthorized" errors: Token may be expired or incorrect
• "Not found" errors: Check enrollment URL and project RID
• Connection timeouts: Verify network access to Palantir
• MCP not available: Ensure platform admin has enabled MCP

Support Resources

• Palantir MCP Documentation
• Palantir Foundry Documentation
• Claude Code Documentation
• Gemini CLI Documentation

Contributing

When contributing to this template:

1. Never commit credentials or tokens
2. Test changes with both AI assistants
3. Update documentation for new features
4. Follow Palantir's best practices
5. Ensure code passes Foundry's quality checks

---

Important: This template requires proper Palantir Foundry access and MCP enablement. Contact your Palantir administrator if you encounter access issues.