Model Context Protocol

Structured arXiv for AI agents

LaTeX-parsed papers with addressable node IDs.

sec:3.2 · eq:1 · fig:2 · tab:1 · thm:1

{
  "mcpServers": {
    "sciencestack": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://sciencestack.ai/api/mcp",
        "--header", "x-api-key:${SCIENCESTACK_API_KEY}"
      ],
      "env": {
        "SCIENCESTACK_API_KEY": "sk_live_your_key_here"
      }
    }
  }
}

Get API Key REST API Docs

5 tools, one connection

All tools support batching for efficiency. Search multiple queries, fetch multiple papers, get multiple nodes — in a single call.

TOOLsearchPapers

Input: ["transformer attention", "self-attention mechanism"]

{
  "results": [
    {
      "arxivId": "1706.03762",
      "title": "Attention Is All You Need",
      "authors": ["Ashish Vaswani", "Noam Shazeer", "..."],
      "tldr": "A new simple network architecture based solely on attention mechanisms...",
      "citationCount": 159306
    },
    ...
  ]
}

Stable node IDs for precise references

Every element in a paper has a stable ID. Fetch exactly what you need — no regex, no guessing.

sec:3.2.1

Sections and subsections

eq:1

Numbered equations

fig:2

Figures with images

tab:1

Tables with data

thm:1

Theorems

Workflow tip: Use getPaperOverview first to discover all node IDs in a paper, then use getNodes to fetch specific content.

Built for AI-native workflows

Whether you're building research tools or adding paper lookup to your IDE.

Research Assistants

Build AI agents that can search, read, and cite papers accurately with stable node IDs.

"Find papers on attention mechanisms and explain equation 1 from the transformer paper"

Literature Review

Automate literature surveys by traversing citation graphs forward and backward.

"What papers cite the original transformer paper? Summarize their contributions."

RAG Pipelines

Ground LLM responses in verifiable paper content with precise section references.

Chunk papers by section, embed with nodeIds for retrieval with attribution.

IDE Integrations

Add paper lookup to Cursor, VS Code, or any MCP-compatible editor.

"Explain this loss function" → Searches papers, returns relevant equations.

MCP vs REST API

Same data, different interface. Use what fits your stack.

Feature	MCP	REST API
Claude Desktop / Cursor integration	✓	—
Tool calling with AI SDK	✓	✓
Direct HTTP requests	—	✓
Batch operations	✓	✓
Same underlying data	✓	✓
Same API key	✓	✓

Not sure which to use? Start with the REST API if you're building a custom integration, or MCP if you want plug-and-play with Claude/Cursor.

Ready to connect?

Get your API key and start building AI-powered research tools.

Get API Key View REST API