DataSF MCP Server

A Model Context Protocol (MCP) server that provides LLMs with seamless access to San Francisco's open data portal (DataSF), powered by the Socrata platform.

Overview

This MCP server enables AI assistants like Claude to search, explore, and query San Francisco's public datasets through a simple, standardized interface. It handles the complexity of the Socrata API, provides intelligent column name correction, and includes schema caching for optimal performance.

Key Features

• 🔍 Dataset Search & Discovery - Find datasets by keywords or browse by category

• 📊 Schema Retrieval - Get column names and data types before querying

• 💬 SoQL Query Execution - Run SQL-like queries against any dataset

• 🎯 Fuzzy Column Matching - Auto-corrects typos in column names

• ⚡ Schema Caching - Reduces API calls with intelligent caching

• 🔐 Optional Authentication - Supports Socrata App Tokens for higher rate limits

• ✅ Property-Based Testing - Comprehensive correctness guarantees

Available Tools

1. search_datasf

Search for datasets by keywords.

Parameters:

• query (string, required): Search keywords (1-500 characters)

• limit (number, optional): Max results (default: 5, max: 20)

Example:

2. list_datasf

Browse available datasets, optionally filtered by category.

Parameters:

• category (string, optional): Filter by category

• limit (number, optional): Max results (default: 5, max: 20)

Example:

3. get_schema

Get the schema (columns and data types) for a specific dataset.

Parameters:

• dataset_id (string, required): Dataset 4x4 ID (format: xxxx-xxxx)

Example:

4. query_datasf

Execute a SoQL (Socrata Query Language) query against a dataset.

Parameters:

• dataset_id (string, required): Dataset 4x4 ID

• soql (string, required): SoQL query (1-4000 characters)

• auto_correct (boolean, optional): Enable column name correction (default: true)

Example:

Installation

Prerequisites

• Node.js 18 or higher

• npm or yarn

Local Setup (Optional)

If you want to run or modify the server locally:

1. Clone the repository:

2. Install dependencies:

3. Run the server:

The server uses tsx to run TypeScript directly without a build step.

Usage

Testing with MCP Inspector

For the MCP Inspector, you'll need to use the local installation:

In the inspector UI, use:

• Command: tsx

• Arguments: src/index.ts (or absolute path if running from outside the directory)

Quick Start with npx (Recommended)

The easiest way to use the server is directly from GitHub using npx:

This will automatically download and run the latest version from GitHub without any manual installation.

Local Installation

Alternatively, clone and install locally:

Then use the absolute path in your MCP configuration (see below).

Configuration for Claude Desktop

Add to your Claude Desktop config file:

Windows: %APPDATA%\Claude\claude_desktop_config.json
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Linux: ~/.config/Claude/claude_desktop_config.json

Option 1: Using npx (recommended)

Option 2: Using local installation

Important: Replace /absolute/path/to/datasf-mcp with the actual full path to where you cloned this project.

Configuration for Kiro IDE

Create or edit .kiro/settings/mcp.json:

Option 1: Using npx from GitHub (recommended)

Option 2: Using local installation

Getting a Socrata App Token

The server works without authentication for public data, but an App Token increases rate limits:

1. Visit https://data.sfgov.org/

2. Sign up for a free account

3. Navigate to Developer Settings

4. Create a new App Token

5. Add it to your MCP configuration

Development

Project Structure

Available Scripts

• npm run build - Compile TypeScript to JavaScript

• npm start - Run the compiled server

• npm test - Run all tests

• npm run test:watch - Run tests in watch mode

Running Tests

The project uses property-based testing with fast-check to ensure correctness across a wide range of inputs.

Architecture

The server follows a modular architecture:

1. MCP Server - Handles protocol communication via stdio

2. Socrata Client - Manages HTTP requests to Socrata APIs

3. Validator - Validates all inputs using Zod schemas

4. Fuzzy Matcher - Corrects column name typos using Fuse.js

5. Schema Cache - Caches dataset schemas in memory (5-minute TTL)

6. Error Handler - Classifies and formats errors for LLM consumption

Example Queries

Once configured in your LLM, you can ask questions like:

• "Search for datasets about housing in San Francisco"

• "What's the schema for the police incidents dataset (wg3w-h783)?"

• "Show me the top 10 incident categories from the police incidents dataset"

• "Find all building permits issued in 2024"

• "What datasets are available about transportation?"

API Endpoints Used

The server interacts with three Socrata APIs:

• Discovery API: https://api.us.socrata.com/api/catalog/v1 - Dataset search and browsing

• Views API: https://data.sfgov.org/api/views/{id}.json - Schema retrieval

• Resource API: https://data.sfgov.org/resource/{id}.json - Data querying

Error Handling

The server provides descriptive error messages for:

• Validation errors - Invalid input format or length

• Not found - Dataset doesn't exist

• Rate limiting - Too many requests (add App Token to resolve)

• Timeouts - Request exceeded 30 seconds

• API errors - Socrata-specific errors (e.g., SoQL syntax errors)

Contributing

Contributions are welcome! The project uses:

• TypeScript for type safety

• Zod for runtime validation

• fast-check for property-based testing

• Vitest as the test runner

License

MIT

Resources

• DataSF Portal

• Socrata API Documentation

• SoQL Query Language

• Model Context Protocol

Troubleshooting

Server not starting

• Ensure you ran npm run build first

• Check that Node.js 18+ is installed

Tools not showing up in LLM

• Verify the path in your config is absolute

• Restart your LLM application after adding the config

• Check the LLM's logs for connection errors

Rate limiting errors

• Add a Socrata App Token to your configuration

• Reduce the frequency of requests

Column name errors in queries

• Use get_schema first to see valid column names

• Enable auto_correct: true (default) for automatic typo correction