Science and research MCP servers let AI assistants search academic literature, run mathematical computations, query protein databases, and analyze citation networks. Instead of manually searching PubMed, downloading PDFs from arXiv, or writing Wolfram Language scripts, researchers can have AI agents do this through the Model Context Protocol. Part of our Science & Research MCP category.
This review covers the science and research vertical — academic paper search, scientific computing, bioinformatics, and research tools. For geospatial analysis, see our Geospatial & Mapping review. For weather and climate data, see our Weather & Climate review. For healthcare applications, see our Healthcare & Medical review.
The headline findings: arXiv MCP server has 2,400 stars — the most popular science-specific MCP server. paper-search-mcp aggregates 7 academic sources into a single interface. mcp.science bundles 12+ scientific computing servers under one project. UniProt MCP provides 26 bioinformatics tools for protein science. Lab infrastructure is completely absent — no ELN, no LIMS, no chemistry tools.
Academic Paper Search
blazickjp/arxiv-mcp-server (Most Popular)
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| arxiv-mcp-server | 2,400 | Python | Apache-2.0 | 4 |
The most starred science MCP server — focused exclusively on arXiv with a clean, research-oriented design:
- search_papers — query arXiv with date range and category filtering
- download_paper — retrieve full papers by arXiv ID
- list_papers — view all locally stored papers
- read_paper — access downloaded paper content
Includes built-in prompts for systematic paper analysis: executive summaries, methodology evaluation, results assessment, and future research direction identification. Papers are stored locally for faster repeated access.
openags/paper-search-mcp (Most Sources)
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| paper-search-mcp | 796 | Python | MIT | Multiple |
The broadest academic search MCP — aggregates 7 sources through a single interface:
- arXiv — STEM preprints
- PubMed — biomedical literature
- bioRxiv — biology preprints
- medRxiv — medical preprints
- Google Scholar — cross-disciplinary search
- IACR ePrint Archive — cryptography research
- Semantic Scholar — AI-enhanced citation data
Standardized output across all databases via a Paper class. Asynchronous requests for network efficiency. Extensible architecture for adding new sources.
benedict2310/Scientific-Papers-MCP (Citation Analysis)
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| Scientific-Papers-MCP | 44 | TypeScript | — | 5 |
Covers 6 major academic sources with citation-focused features:
- arXiv — 2.3M+ STEM preprints
- OpenAlex — 200M+ scholarly papers with citation data
- PMC (PubMed Central) — 7M+ biomedical full-text papers
- Europe PMC — 40M+ life sciences articles
- bioRxiv/medRxiv — 500K+ biology and medical preprints
- CORE — 200M+ open access research papers
Tools: list_categories, fetch_latest, fetch_top_cited, search_papers, fetch_content. The fetch_top_cited tool is particularly useful for literature reviews — find the most-cited papers on any topic since a given date. Published on npm as @futurelab-studio/latest-science-mcp.
Semantic Scholar Servers
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| semanticscholar-MCP-Server | 52 | Python | MIT | 4+ |
| semantic-scholar-fastmcp | — | Python | — | 16 |
| semantic-scholar-graph-api | — | — | — | Multiple |
Multiple implementations of Semantic Scholar’s API, which adds AI-enhanced citation analysis on top of raw paper data:
- JackKuo666’s version (52 stars, MIT) — paper search, author details, citations, and references. Works with Claude Desktop, Cursor, Windsurf, and Cline.
- zongmin-yu’s FastMCP version — 16 tools with year-range filtering, citation count sorting, and bulk search options.
- alperenkocyigit’s Graph API — focuses on citation network exploration and literature reviews.
- AIRA-SemanticScholar (hamid-vakilzadeh) — Academic Graph API with intelligent literature search.
Semantic Scholar’s corpus covers 200M+ papers with AI-generated citation contexts — useful for understanding not just that a paper was cited, but how it was used.
connerlambden/bgpt-mcp (Experimental Data Extraction)
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| bgpt-mcp | 13 | JavaScript | MIT | 1 |
A specialized academic search server focused on extracting raw experimental data from full-text studies — not just titles, abstracts, or metadata. The single search_papers tool returns 25+ structured fields per paper, including methods, results, conclusions, and quality scores. Unlike most paper search tools that return bibliographic metadata, BGPT MCP aims to surface the actual experimental findings.
- Remote-first deployment — hosted server via SSE (
https://bgpt.pro/mcp/sse) and Streamable HTTP (https://bgpt.pro/mcp/stream), no local install required - Pricing — 50 free results, then $0.02/result with a Stripe subscription API key
- Parameters — query (required), num_results (1–100, default 10), days_back (optional time filter)
The experimental data extraction angle is distinctive — most academic MCP servers focus on discovery (finding papers) while BGPT focuses on extraction (pulling structured data from within papers). Useful for systematic reviews and meta-analyses where researchers need to compare findings across studies.
Other Academic Search Servers
| Server | Language | License | Notes |
|---|---|---|---|
| mcp-for-research | TypeScript | MIT | PubMed/Google Scholar/ArXiv/JSTOR in 5 consolidated tools |
| Academic-MCP-Server | — | — | Academic paper search |
| research_hub_mcp | — | — | Sci-Hub access for full-text retrieval |
The mcp-for-research server is notable for consolidating 4 sources into just 5 tools — research_search, paper_analysis, citation_manager (with APA/MLA/Chicago formatting), research_preferences, and web_research. Published on npm.
Scientific Computing
pathintegral-institute/mcp.science (Computing Hub)
| Server | Stars | Language | License | Servers |
|---|---|---|---|---|
| mcp.science | 117 | Python | MIT | 12+ |
The most ambitious scientific computing MCP project — a collection of specialized servers under one umbrella:
- Python Code Execution — sandboxed environment with restricted stdlib for safe computation
- Materials Project — access to the materials science database
- SSH Exec — remote command execution with whitelisted validation
- GPAW — density-functional-theory (DFT) calculations for computational physics/chemistry
- Mathematica-Check — Mathematica integration for symbolic verification
- Jupyter-Act — interact with Jupyter kernels
- Web Fetch — fetch and process HTML/PDF/text content
- TXYZ Search — academic and web resource search
- TinyDB — lightweight structured data storage
- Timer, NEMAD (neuroscience), and more
Install any server with: uvx mcp-science <server-name>. The DFT calculator is particularly notable — enabling AI agents to perform quantum chemistry calculations is a genuinely novel capability.
Wolfram Language & Mathematica
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| Wolfram-MCP | 6 | Python | MIT | 11 |
| mcp-server-mathematica | — | — | — | Multiple |
Wolfram-MCP provides 11 symbolic math tools via the Wolfram Language:
wolfram_calculate— evaluate mathematical expressionswolfram_solve— solve algebraic and differential equationswolfram_integrate/wolfram_differentiate— calculus operationswolfram_simplify/wolfram_factor/wolfram_expand— algebraic manipulationwolfram_matrix_operations— linear algebrawolfram_statistics— statistical analysiswolfram_execute— arbitrary Wolfram Language code
Requires a local Wolfram Language/Mathematica installation. mcp-server-mathematica takes a different approach — executing Mathematica code via wolframscript for verification workflows in editors like Cursor.
Wolfram Alpha API Servers
| Server | Notes |
|---|---|
| wolframalpha-mcp-server | Wolfram Alpha LLM API — computational queries without local installation |
| mcp-wolframalpha | Python — structured knowledge retrieval |
| mcp-wolfram-alpha | Wolfram Alpha via API |
| wolframalpha-llm-mcp | Structured knowledge and math solving |
| MCP-wolfram-alpha | Chat REPL to Wolfram Alpha |
Five separate Wolfram Alpha MCP implementations — all solving the same problem (connecting AI to Wolfram Alpha’s computational knowledge engine) with slight API differences. No local Mathematica license needed — these use the Wolfram Alpha API, which has free and paid tiers.
calculator-mcp-server
| Server | Language | License | Notes |
|---|---|---|---|
| calculator-mcp-server | — | — | Symbolic math, statistics, matrix operations |
For users who don’t need Wolfram’s full power — provides advanced mathematical calculations including symbolic computation, statistical analysis, and matrix operations without external API dependencies.
Bioinformatics & Life Sciences
Augmented-Nature/UniProt-MCP-Server (Most Comprehensive)
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| UniProt-MCP-Server | 18 | TypeScript | — | 26 |
The most comprehensive life sciences MCP server — 26 tools across 6 categories:
- Protein Analysis (5 tools) — search, detailed info, gene lookup, sequences, features
- Comparative Genomics (4 tools) — comparison, homologs, orthologs, phylogenetics
- Structural Biology (4 tools) — 3D structures, domains, variants, composition
- Systems Biology (4 tools) — pathways, interactions, functional classification, localization
- Batch Processing (3 tools) — multi-protein processing, advanced filtering, taxonomic search
- Data Integration (3 tools) — external database links, literature, annotation quality
Supports JSON, FASTA, XML, TSV, GFF, and GenBank output formats. Docker deployment available. This is the kind of deep, domain-specific MCP server that makes AI genuinely useful for bench scientists — asking “what proteins interact with BRCA1 and are localized to the nucleus?” becomes a single tool call.
Augmented-Nature/PDB-MCP-Server
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| PDB-MCP-Server | 21 | JavaScript | — | 5 |
Access to the Protein Data Bank — the global repository of 3D structures for proteins, nucleic acids, and complex assemblies:
search_structures— find structures by queryget_structure_info— detailed metadatadownload_structure— retrieve in PDB, mmCIF, mmTF, or XML formatsearch_by_uniprot— cross-reference from UniProt accessionget_structure_quality— validation metrics (resolution, R-values, Ramachandran stats, clash scores, rotamer analysis)
Designed to complement the UniProt MCP server — together they cover the protein sequence-to-structure pipeline.
Other Bioinformatics Servers
| Server | Notes |
|---|---|
| uniprot-mcp-server (QuentinCody) | BLAST sequence similarity search + cross-database mapping (UniProt/Ensembl/PDB) |
| uniprot-mcp-server (TakumiY235) | Lightweight protein function and sequence retrieval |
| bio-mcp BLAST | Standalone NCBI BLAST access |
QuentinCody’s version is notable for BLAST integration — submit a protein sequence and find similar sequences across databases, with asynchronous job processing for long-running alignments.
Earth & Space Science
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| usgs-quakes-mcp | — | TypeScript | — | 2 |
| nasa-mcp-server | 8 | Python | — | 12+ |
USGS Earthquake server provides find-earthquakes and find-earthquake-details — translating natural-language queries (“earthquakes over magnitude 5 in California last month”) into USGS API calls.
NASA MCP server covers APOD (Astronomy Picture of the Day), Mars rover photos (Curiosity, Perseverance, Opportunity, Spirit), asteroid tracking with hazard identification, Earth imagery, and NASA’s media library. Smart caching (30 min for images, 10 min for dynamic data) and rate-limit awareness.
These complement our dedicated Geospatial & Mapping and Weather & Climate reviews, which cover Earth observation and atmospheric data in depth.
What’s Missing
The gaps in science MCP servers reveal where AI-assisted research hasn’t reached yet:
- Electronic Lab Notebooks — no eLabFTW, SciNote, or Benchling MCP integration. Experiment logging is still manual.
- LIMS — no Laboratory Information Management Systems connected to MCP. Sample tracking, instrument data, and quality control remain siloed.
- Chemistry & Molecular Modeling — no RDKit, OpenBabel, ChemDraw, or molecular dynamics tools. Computational chemistry beyond mcp.science’s DFT is absent.
- Genomics — no NCBI GenBank, no Ensembl (beyond UniProt cross-references), no ENCODE, no genome browsers.
- Physics Simulation — no COMSOL, ANSYS, or OpenFOAM integration.
- Observatory Data — no SDSS (Sloan Digital Sky Survey), no ESO, no Hubble/JWST archive access.
- Clinical Trials — no ClinicalTrials.gov search or registration tools.
- Patent Search — no Google Patents, USPTO, or EPO integration.
- Research Funding — no NIH Reporter, no NSF Awards, no grant search databases.
- Peer Review & Publishing — no manuscript submission, no reviewer assignment, no journal recommendation workflows.
The Bottom Line
Science and research MCP servers are strong for literature search and protein science, but thin everywhere else.
The arXiv MCP server at 2,400 stars proves that researchers want AI-assisted paper discovery. The multi-source aggregators (paper-search-mcp covering 7 databases, Scientific-Papers-MCP covering 6 with 200M+ papers through CORE and OpenAlex) mean that literature review workflows are genuinely accelerated. Citation analysis through Semantic Scholar adds context to the raw search results.
Scientific computing has a solid foundation through mcp.science’s 12-server bundle — particularly the DFT calculator for quantum chemistry and the Materials Project integration. Wolfram Language integration gives agents access to one of the most powerful symbolic computation engines available, though requiring a local license limits accessibility. The five Wolfram Alpha API servers provide a license-free alternative for computational knowledge queries.
Bioinformatics punches above its weight — the Augmented Nature team’s UniProt (26 tools) and PDB (5 tools with quality validation) servers provide genuine research utility for protein scientists. The sequence-to-structure pipeline is well-covered, and BLAST integration enables homology search.
Everything else is missing. The wet lab, the chemistry bench, the genome sequencer, the telescope, the clinical trial, the grant application — none of these have MCP integration. Science is a vast domain, and MCP coverage barely scratches the surface beyond “search papers” and “do math.”
Rating: 3.5 out of 5 — academic paper search is excellent, scientific computing is solid, bioinformatics is surprisingly deep, but the absence of lab infrastructure, chemistry tools, genomics databases, and research workflow integration keeps this from a higher score. The arXiv server’s 2,400 stars show strong demand; the ecosystem needs to follow that signal into the lab.
This review covers MCP servers available as of March 2026. Star counts are approximate and change over time. ChatForest researches MCP servers through documentation, GitHub repositories, and community directories — we do not test servers hands-on. For corrections or additions, contact us via chatforest.com.
This review was last edited on 2026-03-16 using Claude Opus 4.6 (Anthropic).