# Usage Examples
Practical examples for common use cases with the Local LLM MCP Server.
## Table of Contents
- [Getting Started](#getting-started)
- [Privacy & Security](#privacy--security)
- [Code Analysis](#code-analysis)
- [Content Processing](#content-processing)
- [Business & Productivity](#business--productivity)
- [Research & Analysis](#research--analysis)
- [Model Management](#model-management)
---
## Getting Started
### 1. Check Available Models
Before using the server, discover what models are loaded:
```
Ask Claude: "What local models are available?"
```
Claude will read the `local://models` resource and show you something like:
```json
{
"models": [
"openai/gpt-oss-20b",
"qwen3-30b-a3b-deepseek-distill-instruct-2507"
],
"count": 2,
"defaultModel": "openai/gpt-oss-20b"
}
```
### 2. Simple Reasoning Task
```
Ask Claude: "Using the local LLM, explain quantum computing in simple terms"
```
Claude uses the `local_reasoning` tool with the default model.
### 3. Set Preferred Model
```
Ask Claude: "Set qwen3-30b-a3b-deepseek-distill-instruct-2507 as the default model"
```
---
## Privacy & Security
### Analyze Sensitive Documents
**Scenario**: You have a confidential medical report to analyze.
```
Ask Claude: "Analyze this medical report for sentiment locally:
Patient reports feeling better after treatment. Some side effects noted but manageable. Overall positive response to medication regimen."
```
Claude will use `private_analysis` with `domain: medical` to ensure proper context.
### Detect PII in Content
**Scenario**: Check if content contains personally identifiable information.
```
Ask Claude: "Scan this email for privacy issues:
Hi John,
Thanks for your inquiry. Please send the documents to our office at 123 Main St, or email them to processing@company.com. You can also call us at (555) 123-4567.
Best,
Sarah Johnson"
```
Response will identify:
- Names (John, Sarah Johnson)
- Email address
- Phone number
- Physical address
### Rewrite with Privacy Protection
**Scenario**: Share customer feedback without exposing identity.
```
Ask Claude: "Rewrite this customer feedback in a professional style with strict privacy protection:
'Hey, I'm Jane from Acme Corp. Called you guys at 555-1234 yesterday about the billing issue. My account jdoe@email.com was overcharged $500!'"
```
Result:
```
"A customer from an enterprise organization contacted support regarding a billing discrepancy. The account in question was incorrectly charged an amount requiring immediate attention and resolution."
```
---
## Code Analysis
### Security Vulnerability Detection
**Scenario**: Review code for security issues before deployment.
```
Ask Claude: "Analyze this Python code for security vulnerabilities:
```python
def process_user_input(username, password):
conn = db.connect()
query = f\"SELECT * FROM users WHERE username='{username}' AND password='{password}'\"
result = conn.execute(query)
return result
```
"
```
Response identifies:
- SQL injection vulnerability
- Plaintext password handling
- Lack of input validation
### Code Quality Assessment
**Scenario**: Get suggestions for improving code quality.
```
Ask Claude: "Review this JavaScript function for quality improvements:
```javascript
function calc(a, b, op) {
if (op == '+') return a+b;
if (op == '-') return a-b;
if (op == '*') return a*b;
if (op == '/') return a/b;
}
```
"
```
Suggestions include:
- Add JSDoc documentation
- Use switch statement
- Add input validation
- Handle division by zero
- Add type checking
### Find Optimization Opportunities
**Scenario**: Optimize performance-critical code.
```
Ask Claude: "Analyze this code for optimization opportunities:
```python
def find_duplicates(numbers):
duplicates = []
for i in range(len(numbers)):
for j in range(i + 1, len(numbers)):
if numbers[i] == numbers[j] and numbers[i] not in duplicates:
duplicates.append(numbers[i])
return duplicates
```
"
```
---
## Content Processing
### Sentiment Analysis with Domain Context
**Scenario**: Analyze financial news sentiment.
```
Ask Claude: "Analyze the sentiment of this financial news locally:
'The company's Q4 earnings exceeded expectations with a 15% increase in revenue. However, rising operational costs and market uncertainties pose challenges for the upcoming quarter.'"
```
Use `domain: financial` for accurate context.
### Extract Entities from Legal Documents
**Scenario**: Extract key entities from a legal contract.
```
Ask Claude: "Extract entities from this contract excerpt:
'This Agreement is entered into on January 15, 2024, between Acme Corporation, a Delaware corporation with offices at 123 Business Ave, and Beta LLC, represented by John Smith, CEO.'"
```
Extracts:
- Organizations: Acme Corporation, Beta LLC
- People: John Smith
- Locations: Delaware, 123 Business Ave
- Dates: January 15, 2024
- Legal terms: Agreement, CEO, corporation
### Classify Content by Topic
**Scenario**: Categorize support tickets.
```
Ask Claude: "Classify this support ticket:
'My account was locked after trying to reset my password. I've tried the forgot password link but haven't received the email. This is urgent as I need access for work.'"
```
Classification:
- Categories: [technical_support, account_access, password_reset]
- Complexity: medium
- Tags: [urgent, authentication, email_delivery]
---
## Business & Productivity
### Complete Email Templates
**Scenario**: Draft a professional response to a customer inquiry.
```
Ask Claude: "Complete this email template with the given context:
Template:
'Dear [CUSTOMER_NAME],
Thank you for your recent [REQUEST_TYPE] regarding [SUBJECT].
We have reviewed your request and [OUTCOME].
Next steps:
1. [STEP_1]
2. [STEP_2]
3. [STEP_3]
Best regards,
[AGENT_NAME]
[TITLE]'
Context:
Customer Alice Johnson requested information about upgrading to our enterprise plan for her team of 50 users. The request was approved, and she needs to complete payment setup."
```
### Summarize Meeting Notes
**Scenario**: Extract key points from a long meeting transcript.
```
Ask Claude: "Extract key points from this meeting:
'Team discussed Q1 goals. Marketing will launch new campaign in February. Development prioritizing mobile app improvements. Budget concerns raised - need to reduce cloud costs by 20%. Action item: Sarah to present cost analysis next week. Decision made to hire two new developers.'"
```
Response:
```json
{
"keyPoints": [
"Q1 goals discussion",
"New marketing campaign launching February",
"Mobile app improvements prioritized",
"20% cloud cost reduction needed"
],
"actionItems": [
"Sarah presents cost analysis next week",
"Hire two new developers"
],
"insights": [
"Budget optimization is a key priority",
"Team expansion planned in development"
]
}
```
### Risk Assessment
**Scenario**: Assess risks for a new project.
```
Ask Claude: "Using the risk assessment template, analyze this scenario:
'We're planning to migrate our on-premise database to a cloud provider. Timeline is 3 months. Current database has 10TB of customer data. Team has limited cloud experience.'"
```
---
## Research & Analysis
### Synthesize Research Findings
**Scenario**: Combine insights from multiple research papers.
```
Ask Claude: "Synthesize these research findings:
Source 1: Study shows remote work increases productivity by 13%
Source 2: Remote workers report 22% higher job satisfaction
Source 3: Companies save average of $11,000/year per remote worker
Source 4: 15% of remote workers report feeling isolated"
```
### Academic Literature Review
**Scenario**: Analyze papers for a systematic review.
```
Ask Claude: "Conduct a literature review on these machine learning papers, identifying common themes and gaps"
```
---
## Model Management
### Switch Models for Different Tasks
**Scenario**: Use a coding-specialized model for code tasks.
```
Conversation with Claude:
You: "I have two models loaded: openai/gpt-oss-20b and qwen3-30b-a3b-deepseek-distill-instruct-2507. Set the Qwen model as default since it's better for coding."
Claude: [Uses set_default_model tool]
You: "Now analyze this Python code for bugs"
Claude: [Uses local_reasoning with Qwen model automatically]
```
### Model-Specific Requests
**Scenario**: Compare responses from different models.
```
You: "Use openai/gpt-oss-20b to explain quantum computing"
[Response 1]
You: "Now use qwen3-30b-a3b-deepseek-distill-instruct-2507 to explain the same thing"
[Response 2 - possibly different explanation style]
```
---
## Advanced Scenarios
### Multi-Step Privacy Workflow
```
You: "I have a confidential document. First scan it for privacy issues, then rewrite it with strict privacy protection, then summarize the rewritten version."
Claude will:
1. Use private_analysis with analysis_type: privacy_scan
2. Use secure_rewrite with privacy_level: strict
3. Use private_analysis with analysis_type: summary
```
### Code Security Pipeline
```
You: "Review this authentication code for security issues, then if you find any, provide an optimized secure version"
Claude will:
1. Use code_analysis with analysis_focus: security
2. If issues found, use local_reasoning to generate secure alternative
3. Use code_analysis again to verify the new code
```
### Document Processing Chain
```
You: "Analyze this legal contract for entities, classify it by type, extract key points, and identify any privacy concerns"
Claude will:
1. Use private_analysis with analysis_type: entities, domain: legal
2. Use private_analysis with analysis_type: classification, domain: legal
3. Use private_analysis with analysis_type: key_points, domain: legal
4. Use private_analysis with analysis_type: privacy_scan
```
---
## Tips & Best Practices
### Performance Optimization
```
For faster responses on simple tasks:
{
"model_params": {
"temperature": 0.1,
"max_tokens": 200
}
}
For creative or complex tasks:
{
"model_params": {
"temperature": 0.8,
"max_tokens": 2000
}
}
```
### Domain Selection
- Use `domain: medical` for healthcare, patient data, clinical notes
- Use `domain: legal` for contracts, compliance, legal analysis
- Use `domain: financial` for market analysis, financial reports, trading
- Use `domain: technical` for software, engineering, IT documentation
- Use `domain: academic` for research papers, scholarly analysis
- Use `domain: general` for everyday content
### Privacy Levels
- **Strict**: Maximum anonymization, all PII removed (for highly sensitive data)
- **Moderate**: Balance privacy and usability (default, recommended)
- **Minimal**: Basic protection only (for less sensitive content)
---
## Troubleshooting Examples
### Model Not Responding
```
You: "Why isn't the local model responding?"
Check:
1. Read local://status to verify LM Studio is online
2. Read local://models to see if models are loaded
3. Ensure LM Studio local server is running
```
### Choosing the Right Tool
| Task | Tool | Example |
|------|------|---------|
| General questions | `local_reasoning` | "Explain blockchain" |
| Detect sentiment | `private_analysis` (sentiment) | Analyze customer feedback |
| Find security issues | `code_analysis` (security) | Review login function |
| Remove PII | `secure_rewrite` (strict) | Anonymize case study |
| Extract data | `private_analysis` (entities) | Get names from email |
| Fill forms | `template_completion` | Complete email template |
