Skip to content

Quickstart

This path starts from a clean machine with Node.js 18+ and Python 3.10+. It installs the base runtime, verifies it, connects Codex, runs a search, creates a library, and opens the browser UI.

1. Install the base runtime

npx scibudy-install --profile base

The installer creates or updates the Scibudy runtime and keeps local state in the app home, normally ~/.research-mcp.

2. Verify the install

scibudy doctor --json

Check these fields first:

  • status: overall readiness.
  • app_home: where Scibudy keeps runtime state.
  • provider_statuses: which search providers are ready.
  • missing_credentials: optional keys that would improve coverage.
  • codex_configured: whether Codex already has the managed MCP block.

3. Connect Codex

scibudy install-codex
codex mcp get research

You should see a stdio MCP server named research. If Codex is not installed on this machine, skip this step and use the CLI directly.

scibudy search "simulation-based calibration" --limit 10

Use --mode biomed for biomedical searches and --sort recent when recency matters more than relevance.

5. Build a local library

scibudy collect "simulation-based calibration" \
  --target-dir ~/Desktop/sbc-library \
  --limit 30

This writes organized metadata, Markdown, BibTeX, and PDF download results when open-access PDFs are available. Paywalled or unavailable PDFs are listed for manual follow-up instead of being fabricated.

6. Ingest and analyze the library

scibudy libraries
scibudy ingest-library <library_id>
scibudy analyze-topic <library_id> calibration
scibudy search-evidence <library_id> "posterior coverage"

Use the library_id returned by scibudy libraries or the collect command.

7. Open the UI

scibudy ui --open

The UI is a local browser manager for libraries, items, analysis settings, context bundles, and generated reports.

8. Ask Codex to run the workflow

After MCP setup, use a prompt like:

Use research_workflow with query="calibration methods in simulation-based inference", mode="general", limit=50, synthesize=true.

For safer automation, ask for dry_run=true first or use quality_mode=fast for a low-cost search and library setup pass.