Open Science Evidence Finder

Open Science Evidence Finder is an Apify Standby MCP server that retrieves scientific metadata from OpenAlex, Crossref, arXiv, and Europe PMC, then normalizes, deduplicates, scores, and stores evidence candidates for LLM source grounding. It exposes MCP tools for literature search and DOI verification; it does not call an LLM.

Ask for "recent papers about urban heat waves and mortality" or paste a DOI such as 10.1038/s41586-020-2649-2 and get structured metadata back: identifiers, title, authors, venue, publication date, open-access status, source provenance, warnings, and heuristic scores. More free tools and studies are available at mr-bridge.com.

What is Open Science Evidence Finder?

Open Science Evidence Finder is a metadata retrieval Actor for evidence discovery, DOI verification, literature discovery, RAG source grounding, and open-access paper discovery. It queries official open APIs and returns normalized EvidenceItem objects in the default dataset, plus a compact run summary in the key-value store.

It is useful when a LLM needs sourced candidate papers before writing a literature summary, but it should not be treated as a systematic review engine or a scientific-quality judge.

Sources used

• OpenAlex - primary source for works, identifiers, authorship, topics, citations, and open-access metadata.
• Crossref - DOI verification and bibliographic enrichment.
• arXiv - recent preprints and category metadata.
• Europe PMC - biomedical and life-science publications, PMIDs, PMCIDs, abstracts, and open-access flags.

Thank you to arXiv for use of its open access interoperability. This project is not endorsed by arXiv.

How to use Open Science Evidence Finder

1. Create a free Apify account.
2. Start this Actor in Standby mode on Apify. The Input tab is mainly for MCP connection setup; the example prompt is used for saved task examples and one-off runs.
3. Connect your LLM client to https://mrbridge--open-science-evidence-finder.apify.actor/mcp?token=YOUR_APIFY_TOKEN.
4. Ask the LLM for literature discovery or DOI verification. The LLM calls the MCP tools with the right arguments.
5. Read returned tool JSON directly in the LLM, or inspect the default dataset plus RUN_SUMMARY and OUTPUT in Apify storage.

MCP connection

The Actor exposes Streamable HTTP MCP at /mcp:

https://mrbridge--open-science-evidence-finder.apify.actor/mcp?token=YOUR_APIFY_TOKEN

Client setup:

• Claude Desktop -> Settings -> Connectors -> Add custom connector -> paste the URL.
• ChatGPT (Plus / Pro / Team / Enterprise) -> Settings -> Connectors -> enable Developer mode -> Add custom connector -> paste the URL.
• Apify Universal MCP -> add https://mcp.apify.com?tools=mrbridge/open-science-evidence-finder to your existing config.
• Other MCP clients -> use /mcp over Streamable HTTP with Authorization: Bearer YOUR_APIFY_TOKEN or the ?token= query parameter.

For conversational MCP usage, your research query or DOI is sent by the connected LLM through the MCP tools. The example prompt exists so saved Actor tasks have a visible configured input for public example pages.

MCP tools

find_scientific_evidence

Finds, normalizes, deduplicates, and scores open scientific metadata for a natural-language research request.

Tool arguments:

{
  "query": "retrieval augmented generation evaluation benchmarks since 2021",
  "maxResults": 10
}

verify_scientific_doi

Verifies one DOI and returns normalized metadata enriched from OpenAlex, Crossref, and Europe PMC.

Tool arguments:

{
  "doi": "10.1038/s41586-020-2649-2"
}

The Actor queries OpenAlex, Crossref, arXiv, and Europe PMC automatically for search. OPENALEX_API_KEY and CROSSREF_MAILTO can be supplied as environment variables when needed.

Standby mode and monetization

This server runs in Apify Standby mode as a hosted Streamable HTTP MCP server. Warm requests are suitable for conversational use; the first request after inactivity may take longer while Apify starts the container.

This Actor uses Apify pay-per-event pricing for successful MCP tool calls. Check the Pricing tab for the live per-tier prices and automatic subscription discounts.

The intended billing events are tool-read for successful find_scientific_evidence calls and tool-analysis for successful verify_scientific_doi calls. Apify may also charge the synthetic apify-actor-start event when the Standby container starts.

Do not enable apify-default-dataset-item; tool calls already charge by MCP event, and EvidenceItems are stored in the dataset for observability. Charging both the tool event and dataset rows would double bill the same user action. To cap spend, set a max cost per run in Apify Console; the MCP tool returns a clear cost-cap message when maxTotalChargeUsd is reached.

Output

Each dataset row is a normalized EvidenceItem:

{
  "itemType": "work",
  "title": "Example scientific work",
  "normalizedTitle": "example scientific work",
  "doi": "10.1000/example",
  "ids": {
    "openalex": "https://openalex.org/W123",
    "doi": "10.1000/example",
    "pmid": null,
    "pmcid": null,
    "arxiv": null,
    "crossref": "10.1000/example"
  },
  "publicationYear": 2024,
  "publicationDate": "2024-01-15",
  "workType": "journal-article",
  "venue": {
    "name": "Example Journal",
    "issn": ["1234-5678"],
    "publisher": "Example Publisher"
  },
  "authors": [
    {
      "name": "A. Researcher",
      "orcid": null,
      "institutions": []
    }
  ],
  "abstract": "Short abstract when available and allowed.",
  "abstractSource": "crossref",
  "abstractTruncated": false,
  "openAccess": {
    "isOpenAccess": true,
    "oaStatus": "gold",
    "url": "https://example.org/work",
    "license": "cc-by"
  },
  "metrics": {
    "citedByCount": 42,
    "referencedWorksCount": null
  },
  "topics": ["information retrieval"],
  "keywords": ["retrieval"],
  "sourceCoverage": {
    "openalex": true,
    "crossref": true,
    "arxiv": false,
    "europepmc": false
  },
  "sourceRecords": [
    {
      "source": "openalex",
      "recordId": "https://openalex.org/W123",
      "recordUrl": "https://openalex.org/W123",
      "apiUrl": "https://api.openalex.org/works?...",
      "retrievedAt": "2026-06-24T00:00:00.000Z"
    }
  ],
  "scores": {
    "relevanceScore": 0.91,
    "evidenceScore": 0.84,
    "recencyScore": 1,
    "metadataCompletenessScore": 0.88
  },
  "warnings": [],
  "raw": null
}

The Actor also writes:

• RUN_SUMMARY - query, DOI, requested/succeeded/failed sources, raw and deduplicated counts, warnings, and up to five top items without raw records.
• OUTPUT - MCP-friendly object containing resultsCount, compact results, and the same summary.

Scoring

Scores are deterministic metadata heuristics between 0 and 1:

• relevanceScore combines source rank, query-term matches in title/abstract/topics/keywords, and exact DOI match when DOI mode is used.
• recencyScore favors recent publications while keeping older works eligible.
• metadataCompletenessScore checks DOI, date, authors, venue, abstract availability, OA URL, identifiers, and provenance.
• evidenceScore combines relevance, completeness, log-scaled citations, source coverage, and recency.

These scores are not measures of scientific quality, causal validity, peer-review rigor, consensus, or medical/legal reliability.

Limits and responsible use

Metadata may be incomplete, stale, duplicated, or inconsistent across sources. This Actor is not a systematic review, not medical advice, and not legal advice.

The Actor does not download PDFs or full text by default and does not store long copyrighted content. Abstracts are handled conservatively: arXiv and Europe PMC abstracts are used when returned by the API, Crossref abstracts are cleaned and truncated when present, and OpenAlex inverted-index abstracts are not reconstructed unless explicitly requested.

The Actor uses official APIs and includes retries for HTTP 429 and 5xx responses. Keep result limits reasonable and respect each source's API terms, rate limits, attribution expectations, and robots or reuse policies where applicable.

FAQ

Can I use this Actor via MCP?

Yes. This Actor is now a Standby MCP server. Connect your LLM client to /mcp; the available tools are find_scientific_evidence and verify_scientific_doi.

Does it call a LLM?

No. It only retrieves, normalizes, deduplicates, scores, and stores source metadata. The consuming LLM should perform the final synthesis.

What happens if one source fails?

The run continues if at least one requested source succeeds. Source failures are logged and included in RUN_SUMMARY.warnings.

Your feedback

Report bugs, source mapping issues, or feature requests in the Actor's Issues tab on Apify.