# Gemini Url Context

> Expand and analyze URLs using Google Gemini 3.8 Flash URL Context API. Efficiently fetches, summarizes, and extracts information from up to 20 URLs per request with grounding metadata and source citations. Use when analysing or summarising specific URLs using Gemini's 1M-token context window.

- Skill: `dreamlab-ai/gemini-url-context` (Agent Skill)
- Install (CLI): `npx skillmds@latest add dreamlab-ai/gemini-url-context`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dreamlab-ai/gemini-url-context/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Integrations & APIs
- Author: dreamlab-ai (https://skillmd.com/u/dreamlab-ai)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dreamlab-ai/gemini-url-context

---


# Gemini URL Context Skill

Use Google's Gemini 3.8 Flash model to expand, summarize, and analyze web
content directly from URLs, via the `gemini-url-context` subcommand of
`agentbox-mcp` — a single Rust `rmcp` binary (`services/agentbox-mcp`) that
also serves the `imagemagick` and `web-summary` skills. It replaces the former
Python FastMCP server one-for-one: same tool names, parameters, and JSON
response shapes.

## When to Use This Skill

- **URL Expansion**: Fetch and summarize content from any URL
- **Multi-URL Analysis**: Compare or synthesize content from up to 20 URLs
- **Document Extraction**: Extract structured data from web pages
- **Research Aggregation**: Gather information from multiple sources
- **Content Grounding**: Get AI responses grounded in specific web sources

## When Not To Use

- For YouTube video transcript extraction -- use the web-summary skill instead (Gemini does not support YouTube)
- For broad web search across many sources -- use the perplexity-research skill instead
- For browser automation, form filling, or interactive page navigation -- use the browser or playwright skills instead
- For summarising content you already have as text -- no URL expansion tool is needed; process it directly
- For pages behind paywalls or requiring authentication -- Gemini URL Context cannot access restricted content

## Architecture

```
┌─────────────────────────────────┐
│  Claude Code / Skill Invocation │
└──────────────┬──────────────────┘
               │ MCP Protocol (stdio)
               ▼
┌─────────────────────────────────┐
│  Gemini URL Context MCP Server  │
│  (agentbox-mcp gemini-url-      │
│   context, Rust rmcp)           │
└──────────────┬──────────────────┘
               │ HTTPS REST API
               ▼
┌─────────────────────────────────┐
│  Google Gemini 3.8 Flash API    │
│  (with url_context tool)        │
└─────────────────────────────────┘
```

## Tools

| Tool | Description |
|------|-------------|
| `expand_url` | Expand and summarize a single URL |
| `expand_urls` | Batch expand multiple URLs (up to 20) |
| `compare_urls` | Compare content from 2+ URLs |
| `extract_from_url` | Extract specific data from URL content |
| `health_check` | Verify API connectivity |

## Examples

```python
# Expand a single URL
expand_url({
    "url": "https://example.com/article",
    "prompt": "Summarize the key points"
})

# Batch expand multiple URLs
expand_urls({
    "urls": [
        "https://example.com/page1",
        "https://example.com/page2"
    ],
    "prompt": "List the main topics from each page"
})

# Compare URLs
compare_urls({
    "urls": [
        "https://site1.com/product",
        "https://site2.com/product"
    ],
    "prompt": "Compare features and pricing"
})

# Extract structured data
extract_from_url({
    "url": "https://example.com/api-docs",
    "schema": {
        "endpoints": "list of API endpoints",
        "auth_method": "authentication method used",
        "rate_limits": "any rate limiting info"
    }
})
```

## Capabilities & Limits

| Feature | Limit |
|---------|-------|
| URLs per request | 20 max |
| Content size per URL | 34 MB max |
| Supported content | Text, Images, PDFs |
| Not supported | Paywalled, YouTube, Google Workspace, Video/Audio |

## Response Metadata

Responses include `urlContextMetadata` with:
- `retrievedUrl`: The URL that was fetched
- `urlRetrievalStatus`: SUCCESS, FAILED, or PARTIAL

## Environment Variables

| Variable | Required | Description |
|----------|----------|-------------|
| `GOOGLE_API_KEY` | Yes | Gemini API key from https://aistudio.google.com/app/apikey — canonical variable, as registered in `skills/mcp.json` |
| `GOOGLE_GEMINI_API_KEY` | No | Legacy fallback, read only if `GOOGLE_API_KEY` is unset |
| `GEMINI_API_KEY` | No | Legacy fallback, read only if neither of the above is set |
| `GEMINI_MODEL` | No | Model override (default: gemini-3.8-flash) |
| `GEMINI_TIMEOUT` | No | Request timeout in seconds (default: 60) |

## Setup

```bash
# Set API key (add to .env or export)
export GOOGLE_API_KEY="your-api-key"

# Or add to your skills .env
echo 'GOOGLE_API_KEY=your-key' >> "$HOME/.claude/skills/.env"
```

## Troubleshooting

**API Key Issues:**
```bash
# Test API key
curl -s "https://generativelanguage.googleapis.com/v1beta/models?key=$GOOGLE_API_KEY" | jq '.models[0].name'
```

**URL Not Retrieved:**
- Check URL is publicly accessible (no paywall)
- Verify URL returns text/HTML/PDF content
- Check content size < 34MB

## Integration with Other Skills

Combine with:
- `web-summary`: For YouTube transcript extraction (Gemini doesn't support YouTube)
- `perplexity-research`: For broader web search before URL expansion
- `playwright`: For JavaScript-rendered pages that need browser execution