Claude Skills MCP Server

"""Tests for lazy document loading.""" import time import pytest from src.claude_skills_mcp.search_engine import SkillSearchEngine from src.claude_skills_mcp.skill_loader import load_from_github @pytest.mark.integration def test_lazy_document_loading(): """Test that documents are fetched on-demand, not at startup.""" # Time the skill loading (should be fast) start = time.time() skills = load_from_github("https://github.com/K-Dense-AI/claude-scientific-skills") load_time = time.time() - start print(f"\nLoad time: {load_time:.2f} seconds") print(f"Loaded {len(skills)} skills") # Should be much faster than before (< 30 seconds vs 60+ seconds) assert load_time < 30, f"Loading took too long: {load_time}s" # Find a skill with documents skill_with_docs = None for skill in skills: if len(skill.documents) > 0: skill_with_docs = skill break assert skill_with_docs is not None, "Should find at least one skill with documents" print(f"\nTesting with skill: {skill_with_docs.name}") print(f" Documents: {len(skill_with_docs.documents)}") # Documents should have metadata but no content initially first_doc_path = list(skill_with_docs.documents.keys())[0] doc_info = skill_with_docs.documents[first_doc_path] print(f" First document: {first_doc_path}") print(f" Type: {doc_info.get('type')}") print(f" Size: {doc_info.get('size')} bytes") print(f" Fetched: {doc_info.get('fetched')}") assert "url" in doc_info, "Should have URL" assert not doc_info.get("fetched"), "Should not be fetched yet" assert "content" not in doc_info, "Should not have content yet" # Now fetch the document print("\n Fetching document on-demand...") content = skill_with_docs.get_document(first_doc_path) assert content is not None, "Should fetch content on-demand" assert content.get("fetched"), "Should be marked as fetched" if doc_info.get("type") == "text": assert "content" in content, "Text document should have content" print(f" ✓ Fetched text document ({len(content.get('content', ''))} chars)") # Verify it's cached in memory print("\n Testing memory cache...") cached_content = skill_with_docs.get_document(first_doc_path) assert cached_content == content, "Should return cached content" print(" ✓ Memory cache working") @pytest.mark.integration def test_document_disk_cache(): """Test that fetched documents are cached to disk.""" from src.claude_skills_mcp.skill_loader import _get_document_cache_dir skills = load_from_github( "https://github.com/K-Dense-AI/claude-scientific-skills", subpath="scientific-thinking", ) print(f"\nLoaded {len(skills)} skills from scientific-thinking") # Find a skill with documents skill_with_docs = None for skill in skills: if len(skill.documents) > 0: skill_with_docs = skill break assert skill_with_docs is not None print(f"\nTesting disk cache with skill: {skill_with_docs.name}") # Get first document doc_path = list(skill_with_docs.documents.keys())[0] print(f" Document: {doc_path}") # Fetch document (should cache to disk) content1 = skill_with_docs.get_document(doc_path) assert content1 is not None # Check cache directory cache_dir = _get_document_cache_dir() cache_files = list(cache_dir.glob("*.cache")) print(f" Cache directory: {cache_dir}") print(f" Cache files: {len(cache_files)}") assert len(cache_files) > 0, "Should have cache files" # Fetch again (should use disk cache, no network) content2 = skill_with_docs.get_document(doc_path) assert content1 == content2, "Should return same content from cache" print(" ✓ Disk cache working") @pytest.mark.integration def test_startup_time_improvement(): """Verify that startup time is significantly improved.""" print("\n" + "=" * 80) print("STARTUP TIME COMPARISON TEST") print("=" * 80) # Test with full scientific skills repository start = time.time() skills = load_from_github("https://github.com/K-Dense-AI/claude-scientific-skills") load_time = time.time() - start print(f"\nFull repository load time: {load_time:.2f} seconds") print(f"Skills loaded: {len(skills)}") # Count total documents total_docs = sum(len(skill.documents) for skill in skills) print(f"Total documents (metadata): {total_docs}") # Verify load time is acceptable assert load_time < 30, f"Load time {load_time:.2f}s exceeds 30s limit" print(f"\n✓ Startup time: {load_time:.2f}s (target: <30s)") print(f"✓ Expected improvement: ~60s → {load_time:.2f}s") print("✓ Documents are lazy-loaded (not fetched at startup)") print("=" * 80) @pytest.mark.integration def test_lazy_fetching_with_server(): """Test lazy fetching through the MCP server interface.""" from src.claude_skills_mcp.server import SkillsMCPServer import asyncio # Load skills (fast - no document content) skills = load_from_github( "https://github.com/K-Dense-AI/claude-scientific-skills", subpath="scientific-thinking/exploratory-data-analysis", ) assert len(skills) >= 1, "Should load EDA skill" # Create server engine = SkillSearchEngine("all-MiniLM-L6-v2") engine.index_skills(skills) server = SkillsMCPServer(engine) # Find EDA skill eda_skill = skills[0] print(f"\nTesting with skill: {eda_skill.name}") print(f" Documents: {len(eda_skill.documents)}") if len(eda_skill.documents) > 0: doc_path = list(eda_skill.documents.keys())[0] # Call read_skill_document (should trigger lazy fetch) result = asyncio.run( server._handle_read_skill_document( { "skill_name": eda_skill.name, "document_path": doc_path, } ) ) assert len(result) == 1 text = result[0].text # Should have fetched and returned content assert "Document:" in text or "Image:" in text print(" ✓ Lazy fetch triggered by read_skill_document") print(" ✓ Document content returned") @pytest.mark.integration def test_metadata_only_at_startup(): """Verify that only metadata is loaded at startup, not content.""" skills = load_from_github( "https://github.com/K-Dense-AI/claude-scientific-skills", subpath="scientific-packages/biopython", ) assert len(skills) >= 1, "Should load biopython skill" biopython = skills[0] print(f"\nSkill: {biopython.name}") print(f"Documents: {len(biopython.documents)}") # Check all documents - none should have content for doc_path, doc_info in biopython.documents.items(): # Should have metadata assert "type" in doc_info, f"{doc_path} missing type" assert "size" in doc_info, f"{doc_path} missing size" assert "url" in doc_info, f"{doc_path} missing URL" # Should NOT have content assert "content" not in doc_info, ( f"{doc_path} should not have content at startup" ) assert not doc_info.get("fetched"), ( f"{doc_path} should not be marked as fetched" ) print(f" ✓ All {len(biopython.documents)} documents have metadata only") print(" ✓ No content loaded at startup (lazy loading working)")