Ieee Research

A Model Context Protocol server that lets LLM agents search academic literature across IEEE Xplore, arXiv, ACM Digital Library (via CrossRef), and a local PDF library of papers you've already downloaded.

Designed to plug into MCP clients like the GitHub Copilot CLI, Claude Desktop, Cursor, Zed, and others.

⚠️ This server only uses official, sanctioned APIs and your local files. It does not bypass paywalls. For paywalled full text you still need a subscription (institutional or personal); the server returns metadata and abstracts, plus URLs/DOIs you can resolve through your own access.

Features

Tool	Source	Auth
search_arxiv	arXiv (Atom API)	none
search_acm	ACM Digital Library metadata via CrossRef (member 320)	none
search_crossref	All of CrossRef	none
search_ieee	IEEE Xplore Metadata API	free API key
search_all	Fan-out across IEEE + arXiv + ACM	optional IEEE key
get_doi_metadata	DOI lookup via CrossRef	none
index_pdf_library	Walks a folder, extracts text from PDFs, caches index	local
search_pdf_library	Substring search with snippets across indexed PDFs	local
read_pdf_text	Extracts text from a single PDF	local

Install

git clone https://github.com/kevinzhao-tech/ieee-research-mcp
cd ieee-research-mcp
uv sync          # installs into .venv
cp .env.example .env
$EDITOR .env     # set IEEE_XPLORE_API_KEY (optional) and CONTACT_EMAIL

Requires Python 3.12+ and uv.

Getting an IEEE Xplore API key

The IEEE Xplore Metadata API is free but gated. Approval typically takes a few business days.

1. Create an account at https://developer.ieee.org/.

2. Click Get a free API key and request Xplore Metadata API access. Briefly describe your use case ("personal research assistant tooling" is fine).

3. Once approved, copy the API key into your .env:

   IEEE_XPLORE_API_KEY=your_key_here
   

Without a key, every other tool still works; search_ieee and the IEEE section of search_all will return a structured error explaining how to set the key.

Running the server

uv run ieee-research-mcp

The server speaks MCP over stdio. You normally won't run it manually — your MCP client will spawn it.

Connecting to MCP clients

GitHub Copilot CLI

copilot mcp add ieee-research \
  --command "uv" \
  --args "--directory" "/absolute/path/to/ieee-research-mcp" "run" "ieee-research-mcp"

Or, in your Copilot CLI MCP config JSON:

{
  "mcpServers": {
    "ieee-research": {
      "command": "uv",
      "args": [
        "--directory", "/absolute/path/to/ieee-research-mcp",
        "run", "ieee-research-mcp"
      ]
    }
  }
}

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "ieee-research": {
      "command": "uv",
      "args": [
        "--directory", "/absolute/path/to/ieee-research-mcp",
        "run", "ieee-research-mcp"
      ]
    }
  }
}

Restart Claude Desktop and the tools will appear under the 🔌 menu.

Anything else

Any MCP client that can launch a stdio server will work — point it at uv --directory <repo> run ieee-research-mcp.

Example agent prompts

• "Find recent IEEE papers on federated learning differential privacy."
• "Index ~/papers/distributed-systems and find passages discussing vector clocks."
• "Look up DOI 10.1145/3650200 and summarize the abstract."
• "Compare what arXiv and ACM say about retrieval-augmented generation in 2024."

Local PDF library

The index_pdf_library tool scans a folder recursively, extracts text from each PDF with pypdf, and writes a cache to <folder>/.pdf_index.json. Re-running skips files whose mtime hasn't changed. search_pdf_library does case-insensitive substring search and returns ranked snippets with surrounding context.

This is intentionally simple and offline — no embeddings, no external service. Ideal for keeping your downloaded subscription papers searchable without re-hitting the publisher.

Development

uv sync --extra dev
uv run pytest -q

Tests use respx to mock HTTP and reportlab (optional dev dep) to build text PDFs for the indexer tests; PDF tests are skipped if reportlab is not present.

Project layout

src/ieee_research_mcp/
├── server.py           # FastMCP server + tool registration
├── config.py           # Env loading (IEEE_XPLORE_API_KEY, CONTACT_EMAIL)
├── models.py           # Normalized Paper dataclass
├── sources/
│   ├── arxiv.py
│   ├── crossref.py     # also covers ACM via member filter
│   └── ieee.py
└── pdfs/
    └── library.py      # index, search, read

Scope and limitations

• Metadata only for IEEE / ACM / CrossRef sources. Full text requires your own subscription and is outside the scope of this server.
• No paywall bypass of any kind. Use your institution's proxy / personal subscription to download PDFs, then index them locally.
• Substring search for the PDF library — not semantic / vector search. Pull requests adding an embedding-backed index are welcome.

License

MIT — see LICENSE.