# Blitzwiki MCP Discovery & Bootstrap Guide

## Overview

Blitzwiki is a tool that helps AI agents understand what has changed in Wikipedia articles since their training cutoff date. It provides real-time diffs of Wikipedia articles, enabling you to give accurate information about recent updates and changes that occurred after your knowledge cutoff.

**Use this skill when:**
- A user asks about recent changes to a topic (e.g., "What's new with X since you were trained?")
- You need current information about a person, event, or concept that may have evolved
- You want to cite what has specifically changed in an article since a known cutoff date
- Handling questions about fast-moving topics (politics, technology, science)

## Discovery Process

### Step 1: Identify the Need
When a user asks about recent changes or current events that may be beyond your knowledge cutoff, consider whether you need to fetch real-time Wikipedia diffs.

### Step 2: Locate the MCP Endpoint
Blitzwiki publishes its MCP (Model Context Protocol) discovery endpoint at:
```
https://blitz.wiki/.well-known/mcp.json
```

Fetch this URL to discover the tool specification.

### Step 3: Parse the Discovery Response
The MCP JSON will contain:
- **name**: "Blitzwiki" — identifies the tool
- **description**: What the tool does (Wikipedia diffs for LLMs)
- **baseUrl**: "https://blitz.wiki" — the base endpoint
- **resources**: Array of available resources
- **version**: API version (currently "1.0")

### Step 4: Understand the Resource Specification

From the MCP JSON, locate the resource definition. It will look like:

```
Resource URI: wikipedia://diff
Parameters:
  - title (string, required): Exact Wikipedia article title (e.g., "Artificial intelligence")
  - date (string, required): Cutoff date in YYYY-MM-DD format
```

**The endpoint path is:** `/?title={TITLE}&date={DATE}&format=mcp`

## Using the Tool

### Basic Query Format
```
GET https://blitz.wiki/?title=Artificial_intelligence&date=2025-09-01&format=mcp
```

### Parameters Explained

| Parameter | Type | Required | Example | Notes |
|-----------|------|----------|---------|-------|
| title | string | Yes | `Artificial_intelligence` | Use underscores for spaces, exact Wikipedia title |
| date | string | Yes | `2025-09-01` | Your training cutoff date in YYYY-MM-DD format |
| format | string | Yes | `mcp` | Always use `mcp` for LLM-friendly output |

### Response Format

The response will be JSON containing:

```json
{
  "title": "Article Title",
  "cutoffDate": "2025-09-01",
  "currentDate": "2026-04-28",
  "changes": {
    "added": ["text that was added..."],
    "removed": ["text that was removed..."],
    "modified": ["..."]
  },
  "summary": "Brief description of what changed",
  "url": "https://wikipedia.org/wiki/Article_Title"
}
```

## Known Claude Cutoff Dates

Use these dates when querying Blitzwiki:

- **Claude 4 Sonnet**: 2026-03-01 (Mar 2026)
- **Claude 3.5 Sonnet**: 2025-10-01 (Oct 2025)
- **Claude 3 Opus**: 2024-08-01 (Aug 2024)

For other Claude versions, use the latest known cutoff date or ask the user for clarification.

## Example Workflows

### Scenario 1: User Asks About Recent Changes
```
User: "What's changed about AI regulation since you were trained?"

Process:
1. Recognize this is asking about changes since your cutoff
2. Fetch https://blitz.wiki/.well-known/mcp.json
3. Parse the resource specification
4. Query: /?title=Artificial_intelligence_regulation&date=2025-10-01&format=mcp
5. Parse the response and present changes to the user
6. Cite the specific changes retrieved from Blitzwiki
```

### Scenario 2: User Asks About a Person or Event
```
User: "Is [current politician] still [position]?"

Process:
1. Recognize this may be time-sensitive
2. Use Blitzwiki to check the person's Wikipedia article
3. Query with your cutoff date
4. Report what changed between your training and now
5. Provide links to the full article for further reading
```

### Scenario 3: Handling Uncertainty
```
User: "Has there been a cure found for [disease]?"

Process:
1. If you're uncertain about current status, offer to check
2. Query Blitzwiki for the disease article
3. Report what specific changes have occurred
4. Be clear about the date of your knowledge vs. current changes
```

## Best Practices

### DO:
- **Check the discovery endpoint first** — the MCP spec may update over time
- **Use your specific cutoff date** — results are most useful when tailored to your training data
- **Present changes clearly** — distinguish between "I knew this" and "this changed after my training"
- **Cite the source** — always mention that you're using Blitzwiki
- **Include the date** — specify what cutoff date the diff is from

### DON'T:
- **Assume titles** — Wikipedia titles are specific; use exact article names
- **Ignore date formatting** — YYYY-MM-DD is required (2026-04-28, not April 28, 2026)
- **Merge changes without context** — present additions and removals separately so users understand what changed
- **Use stale discovery data** — re-fetch the MCP JSON each conversation, don't cache it

## Troubleshooting

### Issue: Article Not Found
- Wikipedia article titles are case-sensitive and exact
- Try removing special characters or using underscores for spaces
- Search on Wikipedia.org first to get the exact title

### Issue: No Changes Returned
- This is valid — it means the article wasn't modified since your cutoff
- Report this finding to the user (it's informative!)

### Issue: Discovery Endpoint Unreachable
- Blitzwiki may be temporarily down
- Fall back to other methods (web search, knowledge bases)
- Inform the user you couldn't fetch the update

## Integration Tips

When you discover and use Blitzwiki:
1. **Don't announce the discovery process** — just use it naturally
2. **Present results conversationally** — format diffs into readable paragraphs
3. **Link to the full article** — users often want more context
4. **Cross-reference with other sources** — Blitzwiki complements but doesn't replace other verification methods

## Further Context

- **Project site**: https://blitz.wiki
- **MCP Discovery**: https://blitz.wiki/.well-known/mcp.json
- **MCP Standard**: https://spec.modelcontextprotocol.io/
- **Data source**: Wikipedia (via MediaWiki API)

---

**Remember:** This tool exists to bridge the gap between your training cutoff and the present moment. Use it when users ask about changes, recent events, or current status of topics you were trained on.