Best Practices

Get the most out of Gemini MCP Tool with these proven practices.

File Selection

Start Specific

Begin with individual files before expanding scope:

# Good progression
@auth.js                    # Start here
@auth.js @user.model.js     # Add related files
@src/auth/*.js              # Expand to module
@src/**/*.js                # Full codebase analysis

Group Related Files

Include configuration with implementation:

# Good
@webpack.config.js @src/index.js  # Config + entry point
@.env @config/*.js                # Environment + config
@schema.sql @models/*.js          # Database + models

# Less effective
@**/*.js                         # Too broad without context

Query Optimization

Be Specific About Intent

# Vague
"analyze this code"

# Specific
"identify performance bottlenecks and suggest optimizations"
"check for SQL injection vulnerabilities"
"explain the authentication flow step by step"

Provide Success Criteria

# Good
"refactor this to be more testable, following SOLID principles"
"optimize for readability, targeting junior developers"
"make this TypeScript-strict compliant"

Token Management

Gemini Model Selection

• Quick tasks: Use Flash (1M tokens)
• Full analysis: Use Pro (2M tokens)
• Simple queries: Use Flash-8B

Efficient File Inclusion

# Inefficient
@node_modules/**/*.js  # Don't include dependencies

# Efficient
@src/**/*.js @package.json  # Source + manifest

Iterative Development

Build on Previous Responses

1. "analyze the architecture"
2. "focus on the authentication module you mentioned"
3. "show me how to implement the improvements"
4. "write tests for the new implementation"

Save Context Between Sessions

# Create a context file
/gemini-cli:analyze @previous-analysis.md @src/new-feature.js 
continue from our last discussion

Error Handling

Include Error Context

# Good
@error.log @src/api.js "getting 500 errors when calling /user endpoint"

# Better
@error.log @src/api.js @models/user.js @.env 
"500 errors on /user endpoint after deployment"

Provide Stack Traces

Always include full error messages and stack traces when debugging.

Code Generation

Specify Requirements Clearly

# Vague
"create a user service"

# Clear
"create a user service with:
- CRUD operations
- input validation
- error handling
- TypeScript types
- Jest tests"

Include Examples

@existing-service.js "create a similar service for products"

Security Reviews

Comprehensive Security Checks

/gemini-cli:analyze @src/**/*.js @package.json @.env.example
- Check for hardcoded secrets
- Review authentication logic
- Identify OWASP vulnerabilities
- Check dependency vulnerabilities
- Review input validation

Performance Optimization

Measure Before Optimizing

@performance-profile.json @src/slow-endpoint.js 
"optimize based on this profiling data"

Consider Trade-offs

"optimize for speed, but maintain readability"
"reduce memory usage without sacrificing features"

Documentation

Context-Aware Documentation

@src/api/*.js @README.md 
"update README with accurate API documentation"

Maintain Consistency

@docs/style-guide.md @src/new-feature.js 
"document following our conventions"

Common Pitfalls to Avoid

1. Over-broad Queries

❌ @**/* "fix all issues" ✅ @src/auth/*.js "fix security issues in authentication"

2. Missing Context

❌ "why doesn't this work?" ✅ @error.log @config.js "why doesn't database connection work?"

3. Ignoring Model Limits

❌ Trying to analyze 5M tokens with Flash model ✅ Using Pro for large codebases, Flash for single files

4. Vague Success Criteria

❌ "make it better" ✅ "improve performance to handle 1000 requests/second"

Workflow Integration

Pre-commit Reviews

alias gemini-review='/gemini-cli:analyze @$(git diff --staged --name-only) review staged changes'

Daily Development

1. Morning: Architecture review
2. Before PR: Code review
3. When stuck: Debugging help
4. End of day: Documentation updates

Advanced Tips

1. Create Analysis Templates

Save common queries for reuse:

# security-check.txt
Check for:
- SQL injection
- XSS vulnerabilities
- Authentication bypasses
- Rate limiting
- Input validation

2. Chain Operations

"First analyze the bug" → 
"Now write a fix" → 
"Create tests for the fix" →
"Update documentation"

3. Learn from Patterns

When Gemini suggests improvements, ask:

"explain why this approach is better"
"show me more examples of this pattern"