ScholarFetch

ScholarFetch

ScholarFetch is a multi-engine academic research environment for:

• terminal-first literature exploration
• MCP-powered agent workflows
• building curated reading lists and exportable research corpora

It combines:

• a rich interactive CLI for humans
• a classic MCP server (stdio)
• a FastMCP server (stdio, sse, streamable-http)

The core idea is simple: start from keywords, DOI, or authors, traverse papers and references, inspect abstracts and full text, save what matters, then export a compact corpus for synthesis.

What ScholarFetch Does

• Searches across multiple scholarly engines in parallel
• Resolves ambiguous author identities and expands author paper lists
• Traverses references as first-class research nodes
• Retrieves abstracts and machine-readable full text when available
• Tracks a saved paper set during an interactive research session
• Exports citations, abstracts, BibTeX, or full-text corpora
• Exposes the same research workflow to MCP agents
• Maintains stateful saved-paper collections inside one MCP session

Engines

• Elsevier (Scopus / Abstract / Article retrieval)
• OpenAlex
• Crossref
• arXiv
• Europe PMC
• Springer Nature (metadata + open access)
• Semantic Scholar (DOI enrichment path)

Installation

git clone https://github.com/laibniz/scholarfetch.git
cd scholarfetch
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
scholarfetch

Console scripts:

• scholarfetch
• scholarfetch-mcp
• scholarfetch-fastmcp

Alternative:

python3 scholarfetch.py

Credentials

ScholarFetch loads provider credentials server-side / client-side from environment.

Default env file:

• .scholarfetch.env

Typical variables:

ELSEVIER_API_KEY=...
ELSEVIER_INSTTOKEN=...
SPRINGER_META_API_KEY=...
SPRINGER_OPENACCESS_API_KEY=...

Notes:

• ELSEVIER_INSTTOKEN is optional
• provider entitlements and rate limits still apply
• MCP tools do not accept API keys in tool arguments

CLI Research Workflow

ScholarFetch CLI is designed for research traversal.

Typical flow:

1. Start from a topic, DOI, or author.
2. Inspect papers.
3. Read abstracts or full text.
4. Expand references.
5. Jump to related authors.
6. Save promising papers.
7. Export a corpus for downstream work.

Example:

/search graph neural networks
/author Albert Einstein
/papers 1 has:abstract
/article 1
/refs 1
/saved
/export fulltext dummy corpus.txt

CLI Features

• Interactive picker with tree navigation
• Breadcrumbs for current research position
• Action bar for OPEN, ABSTRACT, TEXT, REFS, and AUTHOR
• Backspace to go to parent node
• Esc to return to prompt
• S to save a paper from paper lists or reference lists
• X to remove from the saved list
• AUTHOR action from a paper now lets you select:
  • a single author
  • ALL AUTHORS
• Reference lists behave like paper lists:
  • open
  • abstract
  • text
  • refs
  • author
• Automatic paper availability hints:
  • abstract availability
  • full-text availability
• Progress feedback for expensive transitions
• Interruptible reference preview building with partial results kept

Core CLI Commands

• /search <keywords|doi|person name>
• /author <name>
• /papers <author name|index> [filters]
• /doi <doi>
• /open <index>
• /abstract <doi|index>
• /article <doi|index>
• /refs <doi|index>
• /ref <index>
• /saved
• /export [format style path ...]
• /import [path]
• /pick [mode]
• /config
• /engines
• /help

Paper Filters

Use with /papers:

• year>=YYYY, year<=YYYY, year=YYYY
• has:abstract, has:doi, has:pdf, has:fulltext
• venue:<text>, title:<text>, doi:<text>

Examples:

/papers 1 year>=2020 has:abstract
/papers 1 has:fulltext
/papers andrea de mauro venue:marketing

Export Modes

ScholarFetch supports four export modes from the saved paper set.

• bib
  • BibTeX for citation managers and bibliographic tooling
• citations
  • citation-only export in harvard, apa, or ieee
• abstracts
  • metadata + abstract for each saved paper
• fulltext
  • metadata + abstract + full text when available
  • optional inclusion of references

This makes ScholarFetch useful as a corpus builder for downstream synthesis agents.

MCP Server

ScholarFetch exposes the same research model through MCP.

Modes:

• Classic MCP (stdio): python3 scholarfetch_mcp.py
• FastMCP stdio: python3 scholarfetch_fastmcp.py --transport stdio
• FastMCP SSE: python3 scholarfetch_fastmcp.py --transport sse --host 127.0.0.1 --port 8000
• FastMCP streamable HTTP: python3 scholarfetch_fastmcp.py --transport streamable-http --host 127.0.0.1 --port 8000 --http-path /mcp

Validation:

python3 scholarfetch_mcp.py --self-test
python3 scholarfetch_fastmcp.py --self-test

Public demo endpoints:

• Web UI: https://huggingface.co/spaces/Laibniz/ScholarFetch_Web
• Public MCP endpoint: https://laibniz-scholarfetch-web.hf.space/mcp/
• MCP Registry listing: io.github.laibniz/scholarfetch

MCP Research Model

The MCP server is designed for agent workflows, not only one-off calls.

An agent can:

1. Search papers
2. Resolve authors
3. Expand to author papers
4. Read abstracts / full text
5. Expand references
6. Save promising papers into a named in-memory reading list
7. Export the reading list as:
   • citations
   • abstracts
   • BibTeX
   • full-text corpus

This lets an agent build a focused research set inside one MCP session and then hand off an export artifact to another synthesis step.

See MCP_SERVER.md for the detailed tool model.

Repository Files

• scholarfetch.py: CLI entrypoint
• scholarfetch_cli.py: core CLI + retrieval logic
• scholarfetch_mcp.py: classic MCP server
• scholarfetch_fastmcp.py: FastMCP server
• MCP_SERVER.md: MCP usage guide
• AGENTS.md: agent-facing workflow guide
• SKILL.md: structured research skill guide
• SKILLS.md: index for agent-facing skill docs
• CONTRIBUTING.md: contributor notes

For Agents

If you are running ScholarFetch from an MCP-compatible system, read:

• AGENTS.md
• SKILL.md
• SKILLS.md

These documents explain how to use ScholarFetch as a literature-research environment rather than as a flat search API.

Contributing

See CONTRIBUTING.md.

Security

See SECURITY.md.

License

MIT License. See LICENSE.