Slug MCP

Why?

UCSC students juggle 6+ different websites daily — PISA for classes, nutrition.sa.ucsc.edu for menus, LibCal for study rooms, campusrec for gym status, Metro for buses, the events calendar. None of them talk to each other.

slug-mcp puts all of these behind a single MCP interface. Your AI assistant can combine data across services to answer questions that no single UCSC website can:

"I have CSE 115A at 4pm in Baskin — what's for dinner nearby after class, and when's the next bus home from Science Hill?"

The assistant checks your class schedule for the room and time, pulls tonight's dining hall menus, cross-references dining hours with your class end time, and gets real-time bus ETAs from the nearest stop. One question, four services, ten seconds.

Try These

These prompts show what's possible when campus services are connected:

Features

Quick Start

Prerequisites: Rust 1.75+, Chrome/Chromium (for authenticated tools only)

git clone git@github.com:pronei/slug-mcp.git
cd slug-mcp
cargo build --release

Client Setup

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "slug-mcp": {
      "command": "/path/to/slug-mcp"
    }
  }
}

Claude Code

claude mcp add slug-mcp /path/to/slug-mcp

ChatGPT Desktop

Open ChatGPT > Settings > Beta Features > enable MCP Servers, then:

Settings > MCP Servers > Add Server > Command-line (stdio)

• Name: slug-mcp
• Command: /path/to/slug-mcp

Any MCP-compatible client

Run /path/to/slug-mcp as a stdio subprocess — no arguments needed.

For remote connections, start the SSE server and point your client at the endpoint (see Running Your Own Server).

Authentication

Most tools work without login. Two tools require UCSC authentication:

• Meal balance — Slug Points / Banana Bucks
• Room booking — reserve study rooms

When you ask for something that needs auth, the assistant will call the login tool, which opens a Chrome window for CruzID + Duo MFA. The session is captured automatically and lasts 8 hours.

For remote/headless servers where a browser can't open, generate a portable token locally:

slug-mcp export-token    # opens browser, prints base64 token

Then pass the token to the authenticate tool on the remote server.

Running Your Own Server

# Start the SSE server
./slug-mcp serve --sse --port 3000

For production, put a reverse proxy in front for TLS:

# Caddyfile
slug-mcp.example.com {
    reverse_proxy localhost:3000
}

The server handles multiple concurrent clients. Each SSE session gets its own isolated auth state via the rmcp session factory. Shared resources (cache, HTTP client) are Arc-wrapped and thread-safe.

Architecture

src/
├── main.rs              # CLI (serve / export-token) + service wiring
├── server.rs            # MCP tool handlers (rmcp macros)
├── cache.rs             # Moka TTL cache
├── auth/
│   ├── mod.rs           # AuthManager, SAML-aware HTTP client
│   ├── browser.rs       # Chrome CDP for Shibboleth + Duo login
│   ├── session.rs       # AES-256-GCM encrypted session storage
│   └── token.rs         # Portable base64 token encode/decode
├── dining/              # Menu scraping, nutrition, hours, balance
├── events/              # Tribe Events REST API client
├── recreation/          # Facility occupancy + schedules
├── library/             # LibCal study room availability + booking
├── academics/           # PISA class search + campus directory
├── classrooms/          # Classroom directory + campus locations
├── transit/             # Santa Cruz Metro real-time predictions
├── tides/               # NOAA CO-OPS tide predictions
├── buoy/                # NDBC realtime2 weather + ocean buoys
├── wave_buoy/           # CDIP/NDBC .spec swell vs wind-wave
├── usgs_water/          # USGS NWIS instantaneous stream conditions
├── biodiversity/        # iNaturalist + eBird species observations
└── air_quality/         # EPA AirNow current AQI by ZIP

Optional API keys

Several Santa Cruz / field-research tools need free registration with the upstream provider. Missing keys produce a friendly "not configured" message instead of erroring.

Each module follows the pattern: scraper.rs (HTTP + HTML parsing) > mod.rs (service layer with caching) > server.rs (MCP tool handler).

License

MIT