Home

scomp-link Wiki

Welcome to the scomp-link wiki — the end-to-end ML toolkit with 26 CLI commands, 39 chart types, and native AI agent integration.

pip install scomp-link

Connect any MCP-compatible agent (Claude, Kiro, Cursor, VS Code) to the hosted server:

{"mcpServers": {"scomp-link": {"url": "https://Euribor512-scomp-link.hf.space/sse"}}}


📦 PyPI	pypi.org/project/scomp-link
🐙 GitHub	github.com/GiacomoSaccaggi/scomp_link
🐳 Docker Hub	hub.docker.com/r/jack15121/scomp-link
🤗 HF Space	huggingface.co/spaces/Euribor512/scomp-link
🔧 Smithery	smithery.ai/servers/giacomosaccaggi/scomp-link
📖 Docs	giacomosaccaggi.github.io/scomp_link

Report Builder MCP tools: 6 new stateful tools (report_create, report_add_section, report_add_text, report_add_table, report_add_chart, report_save) for building custom branded HTML reports step-by-step
Configuration system: scomp-link init-config creates persistent branding defaults (~/.scomp-link/config.yaml global, .scomp-link.yaml local). Agents automatically use corporate colors, logo, and footer.
Footer parametrization: footer_html parameter in ScompLinkHTMLReport for fully custom report footers
report_add_chart: Supports all 39 chart types across 3 engines (plotly, rawgraphs, highcharts) with comprehensive agent-friendly documentation
Quick Setup Prompt: Copy-paste prompt in README for instant agent configuration
Stats: 22 MCP tools, 26 CLI commands, 450 tests, 74% coverage

Contrastive Embedding System: EmbeddingSelector auto-selects the best pretrained backbone; fit_head() adds weak learner classifiers on frozen embeddings
Performance: FeatureEngineer migrated to Polars (3-10x faster); DriftDetector and DataQualityReport parallelized with ThreadPoolExecutor
CLI: New embed command for extracting embeddings; text command gains --head, --backbone, --select-backbone
MCP: 16 tools (added embed_text, select_backbone)
InfoNCELoss: Temperature-scaled NT-Xent loss for multi-class contrastive learning
HTML Reports: Contrastive classifier generates validation report with confusion matrix, per-class F1, t-SNE visualization
Coverage: 81% (tests + examples), 349+ tests, 36 examples all passing
Smithery: Published as MCPB bundle + remote URL