Scibudy

Scibudy is a Codex-native scientific research assistant for scholarly search, reusable paper libraries, full-text analysis, and local semantic evidence retrieval.

It is useful when you want Codex or a shell workflow to collect papers, organize a local corpus, ingest PDFs and article pages, search evidence inside that corpus, and generate structured research notes without scattering state across the source checkout.

Scibudy is not a reference manager replacement, a guaranteed full-text downloader for paywalled articles, or a substitute for reading and verifying cited papers yourself.

中文简介：

Scibudy 是一个面向 Codex 的科研增强助手，提供学术检索、文献库管理、全文分析、本地证据检索和期刊文风分析能力。它既可以作为 MCP 工具，也可以作为独立 CLI 和本地管理界面使用。

Status

• License: Apache-2.0
• Current release: v0.3.1
• Release posture: stable v0.x user workflows with explicitly documented limits
• Primary platforms: Linux and macOS
• Full local GPU path: Linux + NVIDIA + conda first

Quick links

• Docs Pages: https://onemule.github.io/SciBudy/
• GitHub Releases: https://github.com/ONEMULE/scibudy/releases
• npm installer: https://www.npmjs.com/package/scibudy-installer
• PyPI package: https://pypi.org/project/scibudy/
• English docs: docs/en/index.md
• 中文文档: docs/zh/index.md
• Prerequisites: docs/en/prerequisites.md / docs/zh/prerequisites.md
• Codex MCP setup: docs/en/codex-setup.md / docs/zh/codex-setup.md
• Usage examples: docs/en/examples.md / docs/zh/examples.md
• Journal style analysis: docs/en/journal-style-analysis.md / docs/zh/journal-style-analysis.md
• Journal text standardization: docs/en/journal-standardization.md / docs/zh/journal-standardization.md
• Architecture: docs/en/architecture.md / docs/zh/architecture.md
• Support matrix: docs/en/support-matrix.md / docs/zh/support-matrix.md
• Contributing: CONTRIBUTING.md
• Security: SECURITY.md
• Support: SUPPORT.md
• Roadmap: ROADMAP.md
• Changelog: CHANGELOG.md

Five-minute start

Requirements for the base path are Node.js 18+ and Python 3.10+.

npx scibudy-install --profile base
scibudy doctor --json
scibudy install-codex
codex mcp get research
scibudy search "simulation-based calibration"
scibudy ui --open

Use scibudy doctor --json to confirm provider readiness, app-home paths, Codex config state, and missing optional credentials.

Install choices

| Path | Best for | Command | | --- | --- | --- | | npm installer | New users and Codex users who want the managed runtime | npx scibudy-install --profile base | | Source install | Contributors or users testing unreleased changes | python -m pip install -e .[dev] then scibudy bootstrap --profile base --install-codex | | GPU local | Linux NVIDIA users who want local embedding and reranking models | npx scibudy-install --profile gpu-local | | Developer install | Repo development, tests, docs, and package checks | python -m pip install -e .[dev,docs,release] |

Profiles:

• base: search, library management, UI, and Codex MCP config
• analysis: base plus analysis-oriented runtime conventions
• gpu-local: local GPU model environment and cache warm flow
• full: base, analysis, and GPU-local setup together

Detailed setup:

• English: docs/en/installation.md
• 中文: docs/zh/installation.md

Common workflows

Search literature

scibudy search "posterior calibration in simulation-based inference" --mode general --limit 20

Build a reusable paper library

scibudy collect "simulation-based calibration" --target-dir ~/Desktop/sbc-library --limit 50
scibudy libraries

Ingest and analyze full text

scibudy ingest-library <library_id>
scibudy analyze-topic <library_id> calibration
scibudy search-evidence <library_id> "posterior coverage"
scibudy synthesize-library <library_id> "calibration in simulation-based inference" --profile auto

Analyze journal writing style

scibudy journal-analyze \
  --journal nature-communications \
  --query "atmospheric chemistry Bayesian inference" \
  --target-dir ./nc-style \
  --target-size 100

Standardize text against a journal corpus

scibudy journal-standardize \
  --corpus-dir ./nc-style \
  --input ./manuscript.tex

Use from Codex

After scibudy install-codex, verify the managed MCP block:

codex mcp get research

Then ask Codex to call the high-level workflow:

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

Use lower-level MCP tools such as search_literature, collect_library, ingest_library, search_library_evidence, and build_research_synthesis when you need manual control.

CLI surfaces

• scibudy
• scibudy-mcp
• Compatibility aliases: research-cli, research-mcp

Examples:

scibudy search "simulation-based calibration" --mode general
scibudy collect "simulation-based calibration" --target-dir ~/Desktop/sbc-library
scibudy journal-analyze --journal nature-communications --query "atmospheric chemistry Bayesian inference" --target-dir ~/Desktop/nc-style
scibudy journal-standardize --corpus-dir ~/Desktop/nc-style --input ~/Desktop/manuscript.tex
scibudy analysis-settings
scibudy ingest-library <library_id>
scibudy search-evidence <library_id> calibration
scibudy profiles
scibudy workflow "calibration methods in simulation-based inference" --limit 50 --topic "calibration in simulation-based inference"
scibudy workflow "calibration methods in simulation-based inference" --dry-run
scibudy workflow "calibration methods in simulation-based inference" --quality-mode fast
scibudy security-audit
scibudy doctor --install-readiness
scibudy synthesize-library <library_id> "causal inference robustness" --profile general
scibudy synthesize-library <library_id> "calibration in simulation-based inference" --profile sbi_calibration
scibudy ui --open

Use dry_run=true when an agent should preview writes and planned steps before executing. Use quality_mode=fast for low-cost exploration, standard for the normal workflow, and deep when missing full text or unsupported claims require stricter follow-up.

For safer agent automation, run scibudy security-audit and scibudy doctor --install-readiness before delegating long-running research workflows.

Domain profiles

Domain profiles do not limit Scibudy's search scope or providers. Search remains general and multi-source by default.

Profiles only tune full-text synthesis: section weighting, evidence markers, unsupported-claim detection, and risk flags.

• general: default all-domain synthesis profile.
• auto: chooses a synthesis profile from the topic while preserving general search.
• sbi_calibration: an example preset for simulation-based inference calibration workflows.

For more examples and Codex prompt patterns:

• English: docs/en/examples.md
• 中文: docs/zh/examples.md

Local model stack

The highest-quality local retrieval path currently uses:

• Qwen/Qwen3-Embedding-4B
• Qwen/Qwen3-Reranker-4B

Recommended workflow:

scibudy install-local-models
scibudy warm-local-models --background

See:

• English: docs/en/gpu-local.md
• 中文: docs/zh/gpu-local.md

Safety and data model

• Runtime state lives in the app home, not in the source directory. The default app home is ~/.research-mcp; override it with RESEARCH_MCP_HOME=/custom/path.
• API keys and provider credentials are written to the app-home .env file. Do not commit that file.
• Source installs and npm installs share the same runtime commands, but generated libraries, caches, reports, and UI state stay outside the repo unless you explicitly choose a repo path.
• GPU-local mode expects a Linux NVIDIA machine with conda. Base and analysis workflows do not require GPU models.
• Scibudy records missing provider credentials and degraded search providers in scibudy doctor --json instead of failing silently.

Repository layout

research_mcp/   Python runtime, MCP server, CLI, analysis engine
web/            UI source and built assets
bin/            npm/bootstrap entrypoints
docs/           Bilingual project documentation
examples/       Copyable usage examples
scripts/        Release and smoke-check helpers
.github/        CI, templates, automation

Open-source project standards

This repository is intentionally organized like a professional open-source library:

• documented install profiles
• release manifest and bootstrap state
• contributor and support policies
• issue/PR templates
• CI and packaging checks
• bilingual documentation for core user workflows

Development

Core local checks:

make test
make build-ui
make package-check
make release-check

For deeper guidance:

• English: docs/en/development.md
• 中文: docs/zh/development.md