PI API MCP Server

A Model Context Protocol (MCP) server that provides standardized tools and resources for interacting with the PI Dashboard API. This implementation enables Claude and other MCP-compatible AI assistants to securely access and manage PI Dashboard resources, including categories and charts.

Utilizing PI with MCP

The following demonstrates typical usage scenarios for this MCP Server after setup completion.

Initial Authentication:

• Execute the following instructions to establish a connection:

Ensure the PI API MCP server is running
Set the API URL to http://localhost:8224/pi/api/v2
Use the authenticate tool for authentication guidance
Check the connection status to verify everything is working
List two charts from the dashboard

Chart Analysis:

• If chart ID 450 contains metadata information, use the following prompt:

Retrieve the metadata from chart ID 450
Extract the chart JSON data from ID 450
Identify chart IDs associated with claims
Obtain JSON data for the identified charts
Analyze the data to generate actionable insights

Example Output:

Installation

Installing via Smithery

To install pi-api-mcp-server for Claude Desktop automatically via Smithery:

npx -y @smithery/cli install @mingzilla/pi-api-mcp-server --client claude

Installation - Using Docker (Recommended)

• No MCP Server configuration needed
• MCP client configuration file setup:

{
  "mcpServers": {
    "pi-api": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "API_URL=http://localhost:8224/pi/api/v2",
        "-e",
        "PI_API_KEY=XXXXXXXX",
        "mingzilla/pi-api-mcp-server"
      ],
      "disabled": false,
      "autoApprove": [
        "keep-session-alive",
        "check-connection",
        "authenticate",
        "list-categories",
        "get-category",
        "list-charts", 
        "get-chart",
        "export-chart",
        "get-filterable-attributes",
        "export-chart"
      ]
    }
  }
}

Important Note: If the --api-url parameter is not provided at initialization, the server will prompt you to configure the API URL using the set-api-url tool before executing any operations. This design enables flexible configuration in environments where the URL is not predetermined at startup.

Configuration File Location

Access your Claude for Desktop application configuration at:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: Use other tools for now. e.g. Cline - ask it to show you the MCP config file

Available Tools

Schema Discovery

• get-filterable-attributes: Get the list of attributes that can be used for filtering by examining a sample entity

  Get the filterable attributes for chart entities
  

Connection Management

• check-connection: Check if the current API URL and authentication are valid
• set-api-url: Configure the base API URL for all requests

  Set the API URL to http://localhost:8224/pi/api/v2
  

Authentication

• authenticate: Get guidance on authentication options
• authenticate-with-credentials: Authenticate with username and password (last resort option)
• keep-session-alive: Verify and refresh the current authentication token (also used for token-based authentication)
• logout: Invalidate the current token and end the session
• set-organization: Set the organization ID for subsequent requests

Categories

• list-categories: List all categories with filtering support
• get-category: Get a category by ID
• create-category: Create a new category
• update-category: Update an existing category
• delete-category: Delete a category
• list-category-objects: List all objects for a specific category

Charts

• list-charts: List all charts with filtering support
• get-chart: Get a chart by ID
• delete-chart: Delete a chart
• export-chart: Export a chart in various formats

Available Resources

• auth://status: Get authentication status
• categories://list: List all categories
• categories://{id}: Get a specific category
• categories://{categoryId}/objects: Get objects for a specific category
• charts://list: List all charts
• charts://{id}: Get a specific chart
• charts://{id}/export/{format}: Export a chart in a specific format

Available Prompts

• analyze-categories: Analyze categories in the dashboard
• analyze-charts: Analyze charts in the dashboard
• compare-charts: Compare data between two charts
• category-usage-analysis: Analyze how categories are being used in charts
• use-filters: Shows how to use filters effectively with this API

Claude Integration Examples

Here are some example queries to use with Claude after connecting the server:

Set the API URL

Please use the set-api-url tool to set the PI API URL to http://localhost:8224/pi/api/v2

Authentication

Please help me authenticate to the PI API.

I have a token. Please use the keep-session-alive tool with my token: [YOUR_TOKEN_HERE]

Please check if my connection to the PI API is working properly.

Working with Categories

List all categories in the dashboard.

Get details about category with ID 123.

Working with Charts

List all the charts available in the dashboard.

Export chart with ID 456 as a PDF.

Using Filters

Get the filterable attributes for chart entities to understand what fields I can filter on.

List charts with description containing "revenue" using the filter option.

Using Analysis Prompts

Analyze the categories in the dashboard.

Compare data between charts 123 and 456.

Show me how to use filters effectively with this API.

---

Development

Local Execution

• Note: you can make use of start.sh to run the dev server as well.

# Clone the repository (SSH or HTTPS option)
git clone git@github.com:mingzilla/pi-api-mcp-server.git
cd pi-api-mcp-server

# Install dependencies
npm install
./dependencies.sh # Installs global dependencies to enable MCP client connection via "@mingzilla/pi-api-mcp-server"

# Build the project
npm run build

# Execute the server
npm start

NPM Installation

# Global installation
npm install -g @mingzilla/pi-api-mcp-server

# Direct execution via npx
npx @mingzilla/pi-api-mcp-server --api-url "http://localhost:8224/pi/api/v2" --auth-token "XXXXXXXX"

MCP Client Configuration

Integration with Claude for Desktop:

Node.js Implementation

• Execute the instructions in the "Local Execution" section
• Ensure ./dependencies.sh has been executed to install required dependencies
• Implement the following configuration (Note: "@mingzilla/pi-api-mcp-server" references the package installed through "Local Execution")

{
  "mcpServers": {
    "pi-api": {
      "command": "npx",
      "args": [
        "-y",
        "@mingzilla/pi-api-mcp-server",
        "--api-url",
        "http://localhost:8224/pi/api/v2",
        "--auth-token",
        "XXXXXXXX"
      ],
      "autoApprove": [
        "keep-session-alive",
        "check-connection",
        "authenticate",
        "list-categories",
        "get-category",
        "list-charts",
        "get-chart",
        "export-chart",
        "get-filterable-attributes",
        "export-chart"
      ]
    }
  }
}

Local Development

• run the server using ./start.sh
• set the config with the path to the build/index.js file

./start.sh

{
  "mcpServers": {
    "pi-api": {
      "command": "node",
      "args": [
        "/home/mingzilla/dev/tool-mcp-pi-api-server/build/index.js",
        "--api-url",
        "http://localhost:8224/pi/api/v2",
        "--auth-token",
        "XXXXXXXX"
      ],
      "autoApprove": [
        "keep-session-alive",
        "check-connection",
        "authenticate",
        "list-categories",
        "get-category",
        "list-charts",
        "get-chart",
        "export-chart",
        "get-filterable-attributes",
        "export-chart"
      ]
    }
  }
}

Development Check List

• update code -> start local server -> test local server with file path to index.js
• update readme.md file -> change the mcpServers config section: docker + node + npx
• ./publish.sh - publish to npm
• ./dockerBuild.sh -> ./dockerPublish.sh (edit version number to match package.json) -> test docker config
• push code to github

License

MIT License

Author

Ming Huang (mingzilla)