Public AI MCP Server

A FastMCP server that provides AI assistants with access to current, community-maintained information and real-time services.

What is Public AI?

Public AI bridges the gap between AI assistants and real-world information. Instead of relying on outdated training data, AI assistants can query this MCP server to access:

• Community-maintained data from wiki.publicai.co
• Real-time APIs for transit, parking, and location services
• Localized resources like crisis hotlines, emergency services, and community information

How It Works

AI Assistant (Claude, etc.)
    ↓ queries
MCP Server (this repository)
    ↓ fetches from
    ├─→ Wiki (community data)
    └─→ External APIs (real-time data)

Understanding Wiki Tools vs MCP Functions

This repository contains MCP functions - Python code that AI assistants can call. These functions access two types of data:

1. Wiki-Based Functions (access community-maintained data)

• Read from wiki.publicai.co where community members maintain Wiki Tools
• Wiki Tools = structured data stored using MediaWiki Cargo (database-like tables)
• Examples: Crisis hotlines, BTO launches, community events
• To contribute data: Edit Wiki Tools on the wiki (no coding required)
• To add new tool types: Create new Wiki Tool with Cargo schema on wiki + add MCP function here

2. API-Based Functions (real-time integrations)

• Call external APIs for live data (transit, parking, maps)
• Examples: Swiss transport, Singapore carpark availability
• To contribute: Add Python code to this repository

Example Flow

1. User asks AI a question: "What's the suicide hotline in Singapore?"
2. AI calls MCP function: use_tool(tool="SuicideHotline", country="Singapore")
3. MCP function queries wiki: Reads from SuicideHotline Wiki Tool's Cargo database
4. AI gets current info: Returns verified, community-maintained resources

Available MCP Functions

Wiki-Based Functions

Functions that read community-maintained Wiki Tools from wiki.publicai.co:

list_tools_by_community(community: str)

List all tools available for a specific community.

Example:

list_tools_by_community(community="Switzerland")
# Returns: List of tools tagged with "Switzerland"

use_tool(tool: str, country: Optional[str] = None, region: Optional[str] = None)

Use a Public AI tool. Automatically adapts based on whether the tool has location-specific resources.

For tools with resources (e.g., SuicideHotline):

use_tool(tool="SuicideHotline", country="Singapore")
# Returns: Crisis hotline numbers and resources for Singapore

For tools without resources (e.g., UpcomingBTO):

use_tool(tool="UpcomingBTO")
# Returns: Full page content about BTO launches

API-Based Functions: Swiss Transport

Real-time Swiss public transport information via transport.opendata.ch:

search_swiss_stations(query: str, limit: int = 10)

Search for train, bus, and tram stations.

search_swiss_stations(query="Zürich HB")

get_swiss_departures(station: str, limit: int = 10)

Get real-time departures with delay information.

get_swiss_departures(station="Bern", limit=5)

plan_swiss_journey(from_station: str, to_station: str, via_station: Optional[str] = None, limit: int = 4)

Plan journeys with real-time connections.

plan_swiss_journey(from_station="Zürich HB", to_station="Geneva")

API-Based Functions: Singapore

Location services for Singapore:

get_singapore_carpark_availability()

Get real-time carpark availability data.

get_singapore_carpark_availability()

API-Based Functions: OpenStreetMap

search_osm_nominatim(query: str, limit: int = 10)

Search for locations worldwide using OpenStreetMap.

search_osm_nominatim(query="Eiffel Tower", limit=5)

Installation

Prerequisites

• Python 3.8+
• FastMCP

Setup

1. Clone the repository:

git clone https://github.com/yourusername/pai-mcp-server.git
cd pai-mcp-server

2. Install dependencies:

pip install -r requirements.txt

3. Run the server:

python main.py

The server will start on http://127.0.0.1:8000.

Configuration

Using with Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "publicai": {
      "command": "python",
      "args": ["/path/to/pai-mcp-server/main.py"]
    }
  }
}

Environment Variables

Currently, no environment variables are required. Future API integrations may require API keys.

Contributing

Public AI has two contribution pathways:

1. Contribute Data (No Coding Required)

Add or update Wiki Tools on wiki.publicai.co:

• Crisis hotline numbers for your country
• BTO launch information
• Community events and resources
• Verify and update existing data

See: Wiki Contribution Guide

2. Add MCP Functions (Python Required)

Integrate new APIs or build functions that require real-time data:

1. Fork this repository
2. Add your Python file to the functions/ folder
3. Test locally
4. Submit a Pull Request

See: CONTRIBUTING.md for detailed guidelines

Example: Adding a New Function

Create a new file in functions/weather.py:

from fastmcp import FastMCP
import json
import urllib.request
from typing import List, Dict, Optional, Any

def register_weather_functions(mcp: FastMCP):
    """Register weather-related MCP functions"""

    @mcp.tool()
    def get_weather_alerts(country: str, region: Optional[str] = None) -> List[Dict[str, Any]]:
        """Get severe weather alerts for a location.

        Args:
            country: Country name (e.g., "Singapore", "Switzerland")
            region: Optional region/state

        Returns:
            List of active weather alerts
        """
        try:
            # Your implementation here
            url = f"https://api.weather.service/alerts?country={country}"
            with urllib.request.urlopen(url, timeout=10) as response:
                data = json.loads(response.read().decode())

            return data.get('alerts', [])
        except Exception as e:
            return [{"error": f"Failed to get alerts: {str(e)}"}]

The function will be automatically loaded by main.py.

Guidelines for MCP Functions

• ✅ Handle errors gracefully
• ✅ Set appropriate timeouts (10s default)
• ✅ Return consistent data structures
• ✅ Document parameters clearly
• ✅ Use environment variables for API keys
• ✅ Create one file per API/service for easy maintenance

Architecture

Wiki-Based Functions Architecture

1. Community edits Wiki Tools on wiki.publicai.co
2. MediaWiki stores data in Cargo tables
3. MCP functions query Cargo API
4. AI assistant gets fresh, community-maintained data

API-Based Functions Architecture

1. AI assistant calls MCP function
2. Function makes API request to external service
3. Function processes and formats response
4. AI assistant gets real-time data

Use Cases

Crisis Support

User: "I'm in Switzerland and need mental health support"
AI: [Calls use_tool(tool="SuicideHotline", country="Switzerland")]
AI: "You can reach Die Dargebotene Hand at 143 (24/7), or via WhatsApp..."

Transit Planning

User: "When's the next train from Zürich to Bern?"
AI: [Calls plan_swiss_journey(from_station="Zürich HB", to_station="Bern")]
AI: "The next train departs at 14:32 from platform 8, arriving at 15:28..."

Community Information

User: "What BTO launches are coming up in Singapore?"
AI: [Calls use_tool(tool="UpcomingBTO")]
AI: "The February 2026 BTO launch includes projects in Bukit Merah, Sembawang..."

Why MCP?

The Model Context Protocol (MCP) allows AI assistants to access tools and data sources in a standardized way. This server:

• Stays current: Community can update wiki data anytime
• Scales easily: Add new tools without retraining AI models
• Community-driven: Non-technical people can contribute data
• Privacy-focused: No user data stored, only serves public information

Technical Details

Dependencies

• FastMCP: MCP server framework
• urllib: HTTP requests (no external dependencies)
• json: Data parsing

API Endpoints Used

• MediaWiki Cargo API: https://wiki.publicai.co/w/api.php
• Swiss Transport: http://transport.opendata.ch/v1/
• OpenStreetMap Nominatim: https://nominatim.openstreetmap.org/

Data Format

All tools return either:

• List[Dict[str, any]]: For list-based results
• Dict[str, any]: For single results

Errors are returned as {"error": "message"} within the response structure.

License

MIT License - See LICENSE file for details

Built for the community, by the community.