Local RAG Search

1---
2name: local-rag-search
3description: Efficiently perform web searches using the mcp-local-rag server with semantic similarity ranking. Use this skill when you need to search the web for current information, research topics across multiple sources, or gather context from the internet without using external APIs. This skill teaches effective use of RAG-based web search with DuckDuckGo, Google, and multi-engine deep research capabilities.
4---
5 
6# Local RAG Search Skill
7 
8This skill enables you to effectively use the mcp-local-rag MCP server for intelligent web searches with semantic ranking. The server performs RAG-like similarity scoring to prioritize the most relevant results without requiring any external APIs.
9 
10## Available Tools
11 
12### 1. `rag_search_ddgs` - DuckDuckGo Search
13Use this for privacy-focused, general web searches.
14 
15**When to use:**
16- User prefers privacy-focused searches
17- General information lookup
18- Default choice for most queries
19 
20**Parameters:**
21- `query`: Natural language search query
22- `num_results`: Initial results to fetch (default: 10)
23- `top_k`: Most relevant results to return (default: 5)
24- `include_urls`: Include source URLs (default: true)
25 
26### 2. `rag_search_google` - Google Search
27Use this for comprehensive, technical, or detailed searches.
28 
29**When to use:**
30- Technical or scientific queries
31- Need comprehensive coverage
32- Searching for specific documentation
33 
34### 3. `deep_research` - Multi-Engine Deep Research
35Use this for comprehensive research across multiple search engines.
36 
37**When to use:**
38- Researching complex topics requiring broad coverage
39- Need diverse perspectives from multiple sources
40- Gathering comprehensive information on a subject
41 
42**Available backends:**
43- `duckduckgo`: Privacy-focused general search
44- `google`: Comprehensive technical results
45- `bing`: Microsoft's search engine
46- `brave`: Privacy-first search
47- `wikipedia`: Encyclopedia/factual content
48- `yahoo`, `yandex`, `mojeek`, `grokipedia`: Alternative engines
49 
50**Default:** `["duckduckgo", "google"]`
51 
52### 4. `deep_research_google` - Google-Only Deep Research
53Shortcut for deep research using only Google.
54 
55### 5. `deep_research_ddgs` - DuckDuckGo-Only Deep Research
56Shortcut for deep research using only DuckDuckGo.
57 
58## Best Practices
59 
60### Query Formulation
611. **Use natural language**: Write queries as questions or descriptive phrases
62   - Good: "latest developments in quantum computing"
63   - Good: "how to implement binary search in Python"
64   - Avoid: Single keywords like "quantum" or "Python"
65 
662. **Be specific**: Include context and details
67   - Good: "React hooks best practices for 2024"
68   - Better: "React useEffect cleanup function best practices"
69 
70### Tool Selection Strategy
71 
721. **Single Topic, Quick Answer** → Use `rag_search_ddgs` or `rag_search_google`
73   ```
74   rag_search_ddgs(
75       query="What is the capital of France?",
76       top_k=3
77   )
78   ```
79 
802. **Technical/Scientific Query** → Use `rag_search_google`
81   ```
82   rag_search_google(
83       query="Docker multi-stage build optimization techniques",
84       num_results=15,
85       top_k=7
86   )
87   ```
88 
893. **Comprehensive Research** → Use `deep_research` with multiple search terms
90   ```
91   deep_research(
92       search_terms=[
93           "machine learning fundamentals",
94           "neural networks architecture",
95           "deep learning best practices 2024"
96       ],
97       backends=["google", "duckduckgo"],
98       top_k_per_term=5
99   )
100   ```
101 
1024. **Factual/Encyclopedia Content** → Use `deep_research` with Wikipedia
103   ```
104   deep_research(
105       search_terms=["World War II timeline", "WWII key battles"],
106       backends=["wikipedia"],
107       num_results_per_term=5
108   )
109   ```
110 
111### Parameter Tuning
112 
113**For quick answers:**
114- `num_results=5-10`, `top_k=3-5`
115 
116**For comprehensive research:**
117- `num_results=15-20`, `top_k=7-10`
118 
119**For deep research:**
120- `num_results_per_term=10-15`, `top_k_per_term=3-5`
121- Use 2-5 related search terms
122- Use 1-3 backends (more = more comprehensive but slower)
123 
124## Workflow Examples
125 
126### Example 1: Current Events
127```
128Task: "What happened at the UN climate summit last week?"
129 
1301. Use rag_search_google for recent news coverage
1312. Set top_k=7 for comprehensive view
1323. Present findings with source URLs
133```
134 
135### Example 2: Technical Deep Dive
136```
137Task: "How do I optimize PostgreSQL queries?"
138 
1391. Use deep_research with multiple specific terms:
140   - "PostgreSQL query optimization techniques"
141   - "PostgreSQL index best practices"
142   - "PostgreSQL EXPLAIN ANALYZE tutorial"
1432. Use backends=["google", "stackoverflow"] if available
1443. Synthesize findings into actionable guide
145```
146 
147### Example 3: Multi-Perspective Research
148```
149Task: "Research the impact of remote work on productivity"
150 
1511. Use deep_research with diverse search terms:
152   - "remote work productivity statistics 2024"
153   - "hybrid work model effectiveness studies"
154   - "work from home challenges research"
1552. Use backends=["google", "duckduckgo"] for broad coverage
1563. Synthesize different perspectives and studies
157```
158 
159## Guidelines
160 
1611. **Always cite sources**: When `include_urls=True`, reference the source URLs in your response
1622. **Verify recency**: Check if the content appears current and relevant
1633. **Cross-reference**: For important facts, use multiple search terms or engines
1644. **Respect privacy**: Use DuckDuckGo for general queries unless specific needs require Google
1655. **Batch related queries**: When researching a topic, create multiple related search terms for deep_research
1666. **Semantic relevance**: Trust the RAG scoring - top results are semantically closest to the query
1677. **Explain your choice**: Briefly mention which tool you're using and why
168 
169## Error Handling
170 
171If a search returns insufficient results:
1721. Try rephrasing the query with different keywords
1732. Switch to a different backend
1743. Increase `num_results` parameter
1754. Use `deep_research` with multiple related search terms
176 
177## Privacy Considerations
178 
179- DuckDuckGo: Privacy-focused, doesn't track users
180- Google: Most comprehensive but tracks searches
181- Recommend DuckDuckGo as default unless user specifically needs Google's coverage
182 
183## Performance Notes
184 
185- First search may be slower (model loading)
186- Subsequent searches are faster (cached models)
187- More backends = more comprehensive but slower
188- Adjust `num_results` and `top_k` based on use case
189