The world's largest open research index integrated in your AI workflows.
Free to start · Usage-based after
Works with
Claude
Cursor
WindsurfUnder two minutes. No coding, no terminal — just paste your key.
Open mcp.sciweave.com in a new tab. Click "Get Your API Key", sign in (or create a free account), and copy the key from your settings page. Free to start.
Open mcp.sciweave.com →Keep this tab open — you'll come back for Step 2.
On a phone? Grab the key now, then finish install on your laptop or desktop — most MCP clients run there.
Pick your client. The instructions below use the exact same API key for all of them.
ChatGPT supports custom MCP connectors via Developer Mode (beta) on Plus, Pro, Business, Enterprise, and Edu plans. Full read + write connectors require Business, Enterprise, or Edu.
ChatGPT's MCP UI changes frequently. Canonical pattern: paste the URL above and your key as the bearer token.
Add to ~/.cursor/mcp.json (or .cursor/mcp.json per project):
{
"mcpServers": {
"sciweave": {
"url": "https://mcp.sciweave.com/mcp",
"headers": { "Authorization": "Bearer YOUR_KEY" }
}
}
}Replace YOUR_KEY, save, and open Cursor Settings → Tools & MCP. Restart if SciWeave doesn't appear.
Add this to ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"sciweave": {
"serverUrl": "https://mcp.sciweave.com/mcp",
"headers": { "Authorization": "Bearer YOUR_KEY" }
}
}
}Replace YOUR_KEY and restart Windsurf.
In any terminal:
claude mcp add --transport http sciweave https://mcp.sciweave.com/mcp \
--header "Authorization: Bearer YOUR_KEY"Replace YOUR_KEY with the key from Step 1. Add --scope user for all projects.
Edit ~/.codex/config.toml (Windows: %USERPROFILE%\.codex\config.toml) and add:
experimental_use_rmcp_client = true
[mcp_servers.sciweave]
url = "https://mcp.sciweave.com/mcp"
http_headers = { Authorization = "Bearer YOUR_KEY" }Replace YOUR_KEY with the key from Step 1, save, and restart Codex. The Codex IDE extension reads the same file — reload the window after editing.
Codex uses --url, not --transport — common failure mode when coming from Claude Code.
In any new chat, try one of the example prompts below. ↓
Five research workflows Claude and ChatGPT simply can't do on their own — until you plug in the SciWeave MCP.
Use ask_research_question to write literature-grounded sections that cite every claim to a verifiable DOI.
"Draft a 300-word background section for a grant on GLP-1 agonists and sustained weight loss. Cover the current mechanistic hypotheses. Cite every claim with a real DOI — I'll verify."
Check novelty and hypothesis gaps without relying on your AI's training data — the MCP queries real papers in real time.
"Has anyone combined graph neural networks with cryo-EM data for de novo protein structure prediction? Cite anything from the last 10 years with a real DOI. If you find nothing, say so explicitly — don't guess."
Generate a curated starter reading list for any subfield — grouped, ranked by impact, every paper verifiable.
"Build me a starter reading list on CRISPR base editing from 2020 onward. Group by topic, rank by citation impact, one sentence on why each paper matters. Every DOI must be real."
Plan keywords, boolean queries, year bounds, and filters — then run an initial sweep grounded in real DOIs.
"Help me design a search strategy for causal evidence on gut microbiome and depression. Propose the key boolean queries, year bounds, and paper filters. Then run an initial sweep and show me the top 5 results with real DOIs."
Pick up research threads from yesterday, last week, or last month. Every citation saved in a thread remains verifiable.
"Resume my research thread on mRNA vaccine cold-chain logistics. Show me where we left off, what was already cited, and what's still open."
RAG eval harness — pipe ask_research_question into your golden-set to benchmark grounding.Long-horizon literature agent — persist thread_id across runs; resume with get_research_thread.Citation-verification step — wrap find_references to post-check any LLM output that mentions a DOI.Three tools, one guarantee — every DOI returned points to a paper that actually exists.
ask_research_questionReturns an AI-generated answer with inline citations grounded in SciWeave's paper index. Supports optional min_year / max_year filters.
ask_research_question(
query="What are the current hypotheses for GLP-1 sustained weight loss?",
min_year=2022
)find_referencesFast reference lookup — returns titles, authors, year, DOI, URL, and abstract snippet for the top top_k most relevant papers. No AI synthesis; ~1 second typical latency.
find_references(
query="CRISPR-Cas9 off-target effects in mammalian cells",
top_k=5
)get_research_threadRetrieves a previous research conversation by its UUID — original question, AI answer, citations, and any follow-ups. Thread IDs are returned by ask_research_question.
get_research_thread(
thread_id="<uuid>"
)What every returned paper includes: title, authors, publication year, DOI, URL, abstract snippet.
ask_research_question → save the returned thread_id → resume later with get_research_thread.
MCP stands for Model Context Protocol — an open standard that lets Claude, ChatGPT, and other AI assistants plug into outside tools and data sources. You care because, without one, the AI is limited to what it memorized during training, which is why it sometimes invents papers or cites the wrong DOI. With the SciWeave MCP installed, Claude can reach into 300 million scientific works in real time and cite the exact ones it used.
Yes — they're two surfaces of the same product. The web app at sciweave.com/web is the full browser experience: sign up, ask questions in a dedicated UI, save papers to personal collections, chat with uploaded PDFs, and run research workflows in a purpose-built interface. The MCP is a second surface that brings the same research index into Claude, ChatGPT, and other AI clients — so you can run the same workflows without leaving the chat you already use. Same account, same index, different doorway. Many researchers use both.
Use the web app if you want a purpose-built research UI with PDF upload, saved collections, and a dedicated workspace for focused research sessions. Use the MCP if you already live inside Claude or ChatGPT and want your research workflows grounded in real papers without tab-switching — or if you're building an agent that needs to call the research index programmatically. Both products use the same account and the same usage-based pricing model.
Claude.ai (web), Claude Desktop, ChatGPT (Plus and Pro, via the Connectors UI), Cursor, Windsurf, and any other client that speaks the Model Context Protocol. Setup takes under two minutes — see the install section above.
Over 300 million scientific works across every discipline — life sciences, physical sciences, social sciences, computer science, humanities. Roughly 200 million of these are peer-reviewed journal articles; the rest are preprints, conference papers, and other scholarly outputs. Updated continuously from the world's largest open research index. Open-access PDF links are included where available.
Title, authors, publication year, DOI, URL, and an abstract snippet. Most papers also come back with novelty signals when SciWeave has computed them.
Your AI's web search returns blog posts, press releases, preprint-mill sites, and whatever is crawlable. It doesn't query a scientific research index and it cannot reliably return real DOIs you can verify — which is why Claude sometimes invents citations when asked for sources. The SciWeave MCP replaces that guessing step with a direct query to 300 million scientific works. Your AI stops inventing citations and starts quoting papers that actually exist, with DOIs you can click.
Open mcp.sciweave.com in a new tab, click "Get Your API Key", sign in (or create a free account — if you already have a SciWeave account for the web app, sign in with that same email, no second account needed), and copy the key from your settings page. Free to start. Then come back here and follow Step 2 of the install section above.
Usage-based. A typical research question costs less than a cup of coffee, and if you don't query, you don't pay anything. There's a free tier to try things out, no credit card required to see the rates, and exact per-query pricing is shown in your dashboard after signup. The web app and the MCP share the same pricing model and the same account — check your SciWeave dashboard for your current credit balance and rates across both surfaces.
Your key is rotatable and revocable from your settings page at any time — if you suspect a leak, generate a new key and the old one stops working immediately. Keys are scoped to the SciWeave MCP only and grant no other access to your account.
Tool-use is deterministic once a connector is installed — Claude calls the MCP for any question that looks like a research query. If you want to be sure on a specific question, just say so: "Use the SciWeave MCP to answer this." Claude will route through the tool and the answer will come back with real citations.
Cancel any time by revoking your API key in settings — there's nothing to unsubscribe from for the MCP. We don't share or sell query data; specifics are in the SciWeave privacy policy.
Every tool listed in the toolbox section above is a standard MCP JSON-RPC endpoint. Add the server once and your agent can call them directly. For literature-grounded reasoning, the canonical loop is: ask_research_question → inspect the returned citations → save the thread_id → resume later with get_research_thread. Three common patterns: (1) RAG eval harness — pipe ask_research_question into your golden-set to benchmark grounding; (2) Long-horizon literature agent — persist thread_id across runs and resume with get_research_thread; (3) Citation-verification step — wrap find_references to post-check any LLM output that mentions a DOI. The SciWeave team is reachable at help@sciweave.com for builder questions.
One API key. Under two minutes to install. Zero fake DOIs.
Questions? Email help@sciweave.com
SciWeave is an AI research assistant developed by DeSci Labs to help users find, analyze, and summarize academic research.
.svg)
© Sciweave 2025 - All Rights Reserved
.svg)
.svg)