Science AI MCP Server

What the SelfPy/science-ai-mcp-server MCP server does

SelfPy/science-ai-mcp-server MCP server brings Science AI Journal workflows into MCP-enabled assistants. It is aimed at researchers and developers building agent-assisted academic workflows, rather than general web search or document storage.

The available functions cover several stages of a publication workflow. A paper can be screened from its title and abstract, checked for possible duplicate publication, matched with journals, and analyzed for research gaps. The server also exposes an AI peer-review pass and controls for the Article Writer pipeline.

How it works

The client launches the package as a local MCP process and passes requests to Science AI Journal using an API key. The configured base URL defaults to the Science AI Journal service and can be changed for staging or self-hosted deployments.

Some operations use local or indexed data rather than an LLM. Paper pre-checking uses FTS5 and returns a predicted acceptance tier from 1 through 5. Duplicate checking searches CrossRef, arXiv, medRxiv, bioRxiv, Unpaywall, and a library containing about 900,000 papers. Journal recommendations use an index of 1,214 venues and include predatory-journal flags.

Other operations consume the caller's Science AI Journal account credits. These include research-gap synthesis, AI review, and Article Writer pipeline operations. The gap tool returns synthesized gaps along with highly cited and recent papers related to a query. The review tool performs a single-agent pass on a prepared prompt.

Setup and configuration

Install and run the SelfPy/science-ai-mcp-server MCP server through npx with Node.js 20 or newer:

{
  "mcpServers": {
    "science-ai-journal": {
      "command": "npx",
      "args": ["-y", "science-ai-mcp-server"],
      "env": {
        "SCIENCE_AI_API_KEY": "saij_your_key"
      }
    }
  }
}

Create the API key from the Science AI Journal account settings. A free account is required, and the token is shown only once. SCIENCE_AI_API_KEY is required. SCIENCE_AI_BASE_URL is optional and defaults to https://scienceaijournal.com; use it when targeting a supported staging or self-hosted endpoint.

The same MCP server configuration is documented for Claude Desktop, Cursor, and Cline. After changing the configuration, fully quit and reopen the host application so it reloads the server. An unauthorized response generally indicates an invalid, revoked, or expired key. The documented rate limit for find_research_gaps is 10 requests per hour per IP.

Tools and capabilities

• pre_check_paper: Estimates a paper's acceptance tier from its title and abstract without an LLM.
• check_duplicate_publication: Cross-references multiple scholarly indexes and the paper library.
• recommend_journals: Ranks possible journals and marks predatory venues.
• find_research_gaps: Produces research gaps and related citation and recency results.
• ai_review_paper: Runs one AI peer-review pass over a prepared prompt.
• start_writer_pipeline and get_writer_pipeline_status: Start and monitor the Article Writer pipeline.

SelfPy/science-ai-mcp-server MCP server is licensed under MIT. Its free tools do not require LLM usage, while credit-based operations depend on the caller's Science AI Journal account balance.