Can I run Pre-Check or the Journal Finder through MCP?

Not yet. The package already registers pre_check_paper and recommend_journals, but the server does not accept API keys on their endpoints yet, so every call gets HTTP 401 and nothing is charged. Run Pre-Check and the Journal Finder in the web app at scienceaijournal.com meanwhile. This page lists them as live once the server accepts API keys there.

Do I need a Science AI Journal account?

Yes — the MCP server sends an API key tied to your account with every tool call, so usage is rate-limited per account and shows up in your activity history. Create a free account at scienceaijournal.com, then generate a key under Settings → API Keys.

What happens when I run out of credits?

A metered call returns HTTP 402 with the credits it needs and the credits you have, and nothing is charged. Top up with a one-time pack of 50 credits at scienceaijournal.com/pricing; the tools that spend no credits keep working either way.

Which MCP clients are supported?

Anything that speaks Model Context Protocol over stdio — Claude Desktop, Claude Code, Cursor, Cline, Continue, the official MCP Inspector, and any custom MCP host. HTTP / SSE transports are deferred until a major host requires them.

What scopes do API keys have?

Every key currently carries the wildcard "*" scope (full access to all MCP-exposed tools). Granular per-tool scopes ship in a follow-up — the database column is a JSON string array so the migration will be zero-downtime.

How do I revoke a key if it leaks?

Go to Settings → API Keys and click Revoke on the affected row. Revocation is immediate (the next request with that token returns 401) and soft — the row stays for audit. Then mint a fresh key and paste it into your MCP client config.

Why don't I see new tools after editing the config?

Restart your MCP client. Claude Desktop reads claude_desktop_config.json only at startup. After restart, look for the 🔌 icon (or equivalent) in the input area to confirm the science-ai-journal server is connected.

Is my paper data sent to Anthropic / OpenAI?

Only when you call an LLM-backed tool — through MCP today that is ai_review_paper — and then only to the provider that powers Science AI Journal (Anthropic Claude). Nothing is sent to OpenAI. find_research_gaps and check_duplicate_publication make no LLM call; the gap search and the duplicate check do look your topic or title up in public scholarly indexes (OpenAlex, CrossRef, PubMed, arXiv and similar) to find matching papers. pre_check_paper and recommend_journals make no LLM call either, once they open to API keys.

Where's the source code?

The MCP server is open source under the MIT licence at github.com/SelfPy/science-ai-mcp-server and is published to npm as science-ai-mcp-server. The routes it calls run on scienceaijournal.com.

Developers · MCP Server · v0.1.3 on npm

Call Science AI Journal tools from Claude Desktop, Cursor & Cline

Install one npm package and our research tools appear inside your MCP-enabled assistant. Search the literature on a topic, check a paper for prior publication, or run a specialist reviewer agent — right where you're working, without switching tabs.

Coming soon via MCP: Pre-Check and the Journal Finder — they run in the web app today.

Generate API key →View on npm Source on GitHub

What it costs

The package itself is free and open source; what each tool call costs is listed below, in credits. A free account starts with 25 welcome credits, and credits come in one-time packs of 50 that never expire.

find_research_gaps: Free search — no credits
check_duplicate_publication: Free — no credits
ai_review_paper: 4 credits per agent call
start_writer_pipeline / get_writer_pipeline_status: Private beta · not via MCP yet
pre_check_paper: Coming soon via MCP · web app today
recommend_journals: Coming soon via MCP · web app today

Not available through MCP yet: the package registers these, but the server turns every API key away on their endpoints for now, and nothing is charged. The catalogue below says why for each.

Quick start (≈ 2 min)

1
Generate an API key
In your Settings → API Keys page, click New API key, name it (e.g. Claude Desktop on laptop), copy the saij_… token. You'll only see it once.
2
Add the MCP server to your client config
Drop this block into your Claude Desktop config file:
```
{
  "mcpServers": {
    "science-ai-journal": {
      "command": "npx",
      "args": ["-y", "science-ai-mcp-server"],
      "env": {
        "SCIENCE_AI_API_KEY": "saij_…paste your key here…"
      }
    }
  }
}
```
Config file location:
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Linux: ~/.config/Claude/claude_desktop_config.json
3
Restart your client and try a tool
Quit and reopen Claude Desktop (or Cursor / Cline). When it comes back, the science-ai-journal server appears in the 🔌 tool list. Try this prompt — the duplicate check spends no credits, so your first call is free:
```
Use science-ai check_duplicate_publication on this paper:
Title: Robust Gradient Estimation for Sparse Cross-Modal Transformers
Abstract: We propose a new gradient-estimation scheme for sparse cross-modal transformers…
```

Tool catalogue

Live today

find_research_gapsFree search — no credits

The free literature search on a topic — the gaps unlock in the web app.

When to use: Early-stage discovery: returns a field overview and the papers the search found — the closest matches from our 4.5M-paper library plus the most-cited and most-recent work on the topic. The research gaps come from unlocking that topic at /research-gaps (50 credits, once — the filled comparison table, the gaps, a research proposal and one Pro Report run are all included), so this tool returns gaps: [] and gapsLocked: true.

Example: “Search the literature on graph neural networks for drug discovery and list the closest papers.”

check_duplicate_publicationFree — no credits

Prior-publication and overlap check before you submit.

When to use: Checks your title (and abstract, to confirm close matches) against 7 scholarly indexes (CrossRef, PubMed, arXiv, bioRxiv, medRxiv, Europe PMC and Unpaywall) and our 4.5M-paper local library, all in parallel with a 12-second cap per source.

Example: “Check my paper for prior publication before I submit — title and abstract are…”

ai_review_paper4 credits per agent call

One specialist reviewer agent on a prepared manuscript prompt.

When to use: Runs a single reviewer agent — methodology, formulas, originality, literature, reproducibility, clarity or figures — and returns a structured verdict: score, decision, strengths, concerns and questions for the authors. Each call is one agent. For a whole manuscript, AI Review in the web app runs all 8 specialist agents plus a synthesis for 30 credits.

Example: “Run a methodology review with ai_review_paper on this prepared prompt: …”

Not available through MCP yet

Registered by the package, but the server turns every API key away on these endpoints for now — nothing is charged. Pre-Check and the Journal Finder run in the web app today.

start_writer_pipeline / get_writer_pipeline_statusPrivate beta · not via MCP yet

Drive the Article Writer pipeline from your assistant.

When to use: Start Step 7.5 (code search, adapt, verify, simulate) or another section of a writer session you already own, then poll its status until it is done.

Today: The Article Writer is in private beta, and during the beta the server answers every API-key call to these two tools with HTTP 404 — writer access is checked against a signed-in browser session, which an API key does not carry. Nothing is charged.

pre_check_paperComing soon via MCP · web app today

Acceptance odds for top, mid and open/emerging journals, from a title and abstract.

When to use: Fast readiness check before submission: the tier probabilities, the detected field, and how many similar papers from our local 4.5M-paper library (FTS5 BM25) the estimate rests on, with no LLM call.

Today: The server does not accept API keys for this tool yet: a call returns HTTP 401 and charges nothing. Once it does, the price is 15 credits per call. Run Pre-Check in the web app →

recommend_journalsComing soon via MCP · web app today

Ranked target-journal shortlist from a 17,500-venue index.

When to use: Up to 25 ranked journals (10 by default) with letter grades, tier, quartile, access status, 2-year citation rate, predatory flags, responding times where the venue reports them, and similar papers from each venue that our library holds. Local BM25 + topic profiles, no LLM call.

Today: The server does not accept API keys for this tool yet: a call returns HTTP 401 and charges nothing. Once it does, the price is 15 credits per call. Use the Journal Finder in the web app →

Other MCP clients

The same npm package works with every MCP host that speaks stdio — only the config format differs.

Claude Code

one command in your terminal

claude mcp add science-ai-journal --env SCIENCE_AI_API_KEY=saij_… -- npx -y science-ai-mcp-server

Cursor

~/.cursor/mcp.json

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

Cline (VS Code)

cline_mcp_settings.json (via the Cline extension panel)

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

Continue / generic

Anywhere that supports an "mcpServers" object

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

Troubleshooting

Tools don't appear after editing config: fully quit the host app (not just close the window) and reopen. On macOS use ⌘Q.

401 from pre_check_paper and recommend_journals: these are not open to API keys yet, so every key gets a 401 there and nothing is charged. The 401 says nothing about your key, even where the error message blames it. Use the web app for now.

401 unauthorized from any other tool: the API key was wrong, revoked, or expired. Generate a fresh one and update the config.

402 insufficient credits: the call costs more credits than the account holds, and nothing was charged. Top up at /pricing; find_research_gaps and check_duplicate_publication keep working meanwhile.

404 from start_writer_pipeline: the Article Writer is in private beta. During the beta the server answers every API-key call to the writer endpoints with 404 — writer access is checked against a signed-in browser session, which an API key does not carry.

429 too many requests / rate-limit on /api/research-gaps: you hit the 10-per-hour IP quota on research gaps. Wait an hour and retry, or move to a different IP.

npx complains it can't resolve the package: ensure Node ≥ 20 and that your npm registry isn't set to a private mirror that hasn't mirrored the package yet. Override with npm config set registry https://registry.npmjs.org/.

Frequently asked questions

The npm package is free and open source (MIT). find_research_gaps (the free literature search) and check_duplicate_publication spend no credits; ai_review_paper costs 4 credits per agent call. A free account starts with 25 welcome credits; after that, credits come in one-time packs of 50 that never expire, with no subscription. pre_check_paper and recommend_journals are not available through MCP yet — a call is turned away and nothing is charged.

Ready to wire it up?

Generate your first key, paste the config, restart your client — two minutes from here to your first check inside Claude Desktop.

Generate API key →