Improve LLM/IDE integration with clearer tool selection guidance

[oscarvalenzuelab]

Summary

Enhances MCP server instructions to improve recognition and usability in LLM-powered development environments (Cursor, Windsurf, etc.).

Problem

When users asked LLM-powered IDEs to "do compliance for this project", the LLM would often:

• Not recognize that mcp-semclone handles compliance tasks
• Attempt to install external tools (npm install license-checker, pip install scancode-toolkit)
• Struggle to select the correct tool from the 14 available options

Root causes:

• No explicit trigger keywords linking compliance terminology to mcp-semclone
• 384-line instruction block potentially exceeded LLM processing windows
• Tool selection required reading through multiple detailed docstrings
• No clear decision tree for common use cases

Solution

1. Added Trigger Keywords (Lines 40-48)

Explicit keywords that help LLMs recognize when to use this server:

• "compliance", "license compliance", "do compliance"
• "SBOM", "software bill of materials", "supply chain security"
• "can I ship this?", "license compatibility"
• "mobile app compliance", "SaaS compliance"

2. Added Decision Tree (Lines 50-76)

Clear IF-THEN logic for tool selection:

IF user says "do compliance" → run_compliance_check()
IF source code directory → scan_directory()
IF package archive → check_package()
IF binary → scan_binary()
IF "can I use [license]?" → validate_policy()

3. Condensed Instructions (384 → 175 lines)

Reduced instruction block by ~55% while maintaining critical information:

• Kept: Anti-patterns, key workflows, tool descriptions, constraints
• Condensed: License interpretation, binary scanning, common workflows
• Removed: Verbose examples, redundant explanations
• Added references to detailed docstrings for deep dives

Changes

• -348 lines / +174 lines in mcp_semclone/server.py
• No functional changes - purely documentation improvements
• All tool implementations remain unchanged

Testing

To verify improvements, test with LLM-powered IDEs:

1. "Do compliance for this project" → Should use run_compliance_check()
2. "Can I ship this with GPL to App Store?" → Should use validate_policy()
3. "Generate SBOM" → Should use generate_sbom()

Expected: LLM immediately recognizes these as mcp-semclone tasks instead of attempting external tool installation.

Impact

• Faster tool recognition for compliance queries
• Reduced hallucination/incorrect tool selection
• Better first-time user experience in LLM IDEs
• Clearer guidance on tool selection without reading full docs