leonjoeeee/papervault
Developer toolsこのページは AI により自動翻訳されています。技術的な詳細は原文をご確認ください。 原文を見る
Industrial-grade, self-hosted literature + knowledge layer for coding agents (Claude Code plugin, MCP underneath)
概要
A self-hosted , delivered as one MCP server. Point any MCP-capable agent (Claude Code, and others) at it and the agent gains three tools: - — discover literature by describing your research intent in prose; relevant finds are auto-ingested (downloaded, OCR'd, full text on disk). - — look up papers by citation key / DOI / arXiv id / fuzzy title; read verbatim full text from a returned path. - — ask the knowledge base a question, get a cited prose answer synthesized from a knowledge graph distilled from everything the library has ingested. The corpus compounds without babysitting: search → queryable knowledge in tens of minutes, self-healing ingest queues, cited answers grounded in real sources. pre-release (closed beta). See docs/PRD.md for scope and docs/architecture.md for structure. This is industrial-grade infrastructure, not a toy — read the hardware floor before installing. papervault runs its own embedding, reranking, and OCR models locally.
README
以下は原文 README の抜粋です。原文を見る。
papervault
A self-hosted literature and knowledge layer for coding agents, delivered as one MCP server. Claude Code, Codex CLI, Antigravity CLI/IDE and other MCP clients share three tools:
| Tool | Purpose |
|---|---|
search_papers |
Discover papers from a natural-language research intent; submit independently admitted registration and background acquisition. |
get_paper |
Resolve a citation key, DOI, arXiv ID or fuzzy title and return metadata plus a serve-safe full-text path when available. |
query |
Retrieve evidence from the knowledge graph and synthesize an answer with verified paper citations. |
Version 0.10.15 expands the default research domain and venue authority to include space computing alongside space physics and AI4Science. The product is pre-release (closed beta). Read the PRD for scope, architecture for boundaries, installation guide for deployment, and baselines for dated measurements. The constitution and refactor plan govern collection integrity and useful retrieval together.
Hardware floor (not optional)
The supported deployment uses Linux and CUDA, one 24 GB-class NVIDIA GPU (RTX
3090/4090 class), Docker for Neo4j and PostgreSQL, and an operator-supplied
OpenAI-compatible LLM endpoint and key. Embedding and reranking run locally; the
independent MinerU OCR service shares the GPU. The hardware floor is part of the
product contract. The factory domain pack is space physics, space computing, and AI4Science;
its search rubric, ontology, examples and instructions live in
src/papervault/domain/. Other domains need their own
configuration and evaluation gold sets.
Space computing covers orbital/satellite edge computing and computing constellations;
on-board AI and payload processing; spaceborne accelerators and radiation/fault tolerance;
space-air-ground computing and inter-satellite offloading/routing/scheduling; and orbital
data centers with compute energy management and thermal control. These additions use the
existing 11 entity types; the bundled pack remains named space_physics.
Quick start
Follow INSTALL.md for the complete setup:
cp .env.example .env # fill in the LLM endpoint, key and data roots
# External Docker volumes hold the backing stores; create once.
docker volume create ks-pgdata
docker volume create ks-neo4jdata
docker compose --env-file .env -f deploy/docker-compose.yml up -d
uv venv && uv pip install -e ".[grey]"
.venv/bin/papervault doctor
.venv/bin/papervault serve
Provision mineru-service separately at http://127.0.0.1:30000 (override with
MINERU_URL). papervault uses the official mineru-api HTTP task API through core
インストール
This server does not publish a one-line install command.
Open the repository installation guide MCP 初心者？mcpskill.org のガイドから始めましょう →