← Back to Autonomy

Research Agent · Source Stack · Field Notes · By Majid Mazouchi

Where to send the agent looking

A plain-language guide to the libraries, code shelves, and lecture halls a scientific research agent should search — and what each one is actually good for.

SRC · PAPERS OpenAlex · S2 · arXiv · DBLP the literature SRC · CODE GitHub (MCP) the implementations SRC · MEDIA YouTube · Hugging Face talks & models DEDUP → RANK → SNOWBALL → CITE cite only what was retrieved — never from memory Literature Survey themes · figures · BibTeX appendix
Fig. 1 — Three streams in, one cited document out. The filter bar is where a real survey is won or lost.

Before the sources, the wiring. Everything on this page hangs off one split: tools retrieve, a skill synthesizes — keep those two layers apart and most of the design follows.

Copilot Agent custom .agent.md · orchestration skill tool calls Papers MCP Code MCP Media MCP People MCP OpenAlex · S2 arXiv · DBLP GitHub YouTube · HF USPTO · ORCID raw results DEDUP · RANK · SNOWBALL · CITE Synthesis skill → Survey / Dossier your house style · BibTeX appendix · cite only what was fetched
Fig. 2 — The wiring. Retrieval lives in the MCP tools; synthesis lives in your skill. The two layers stay apart.

↓ build it: harness kit (.zip)  ·  ↓ mcp-builder skill (.zip)

01

Papers — the heart of a survey

This is where most of the work lives. The trick is that no single database is best at everything, so the agent uses a few together: one for coverage, one for ranking, one for fresh preprints, one for clean citations.

SRC-01 · OpenAlex

The wide net

Casts the broadest possible net so nothing important slips through.

It is a giant open catalogue of scholarly work — over 250 million records — and it is completely free with no API key required. Start here to make sure the survey is complete, then narrow down.

Use it for: coverage, the first broad sweep, citation counts.

Practical: put your email in the request (the "polite pool") for faster, more reliable responses. Generous free limit, ~100k calls/day.

SRC-02 · Semantic Scholar

The smart librarian

Knows not just what exists, but what's worth reading and what connects to what.

Around 200 million papers, each with a one-line AI summary (TLDR), DOIs, arXiv IDs and PDF links. Its real superpower is two endpoints: recommendations (give it a good paper, get similar ones) and the citation graph (what this paper cites, and who cites it).

Use it for: ranking by relevance, and "snowballing" — growing a reading list outward from a few seed papers.

Practical: works without a key, but grab a free API key to lift the rate limit before doing a big survey.

SRC-03 · arXiv + DBLP

Fresh ink & clean catalogue cards

The newest preprints, plus tidy citation records for computer science.

arXiv is where new work appears first — before peer review — so it keeps the survey current. DBLP is the meticulous index for CS and engineering venues and hands back accurate, ready-to-use BibTeX.

Use it for: the latest preprints (arXiv) and correct author/venue/citation formatting (DBLP).

Practical: arXiv asks you to throttle — about one request every 3 seconds. Both are free, no key.

SRC-04 · IEEE Xplore

The locked vault

Where much of the control and power-electronics literature actually lives.

For your field — motor control, inverters, FOC, diagnostics — a lot of the canonical work is in IEEE journals and conferences. But the Xplore API is metadata-only and access-restricted: you can find papers, not freely download them.

Use it for: confirming the official version of record in your domain.

Practical: you'll want GM library / institutional access for full PDFs. Workflow: discover via OpenAlex or S2, then pull the official copy through IEEE.

Papers — at a glance
SourceBest atKey neededFull text
OpenAlexCoverage, breadthNoOpen-access links
Semantic ScholarRanking, recommendations, citation graphOptionalPDF links where OA
arXivLatest preprintsNoYes (free)
DBLPClean CS BibTeXNoMetadata only
IEEE XploreDomain venues of recordYesRestricted
02

Code — what people actually built

A paper tells you the idea; the repository tells you whether it runs. The official tool here is the GitHub MCP server, which lets the agent search code and read READMEs directly.

SRC-05 · GitHub (MCP)

The code shelf

Finds the implementations behind the ideas — but only if you ask the right way.

The catch in plain words: GitHub search matches keywords, not meaning. If you type the exact phrase from a paper and it isn't in the repo's text, you get nothing — even when a perfect implementation exists under a different name.

So the agent must: expand the query with synonyms and acronyms (e.g. "field-oriented control", "FOC", "vector control"), then rank results by stars + how recently it was committed to + how well the README matches — not by GitHub's relevance score alone.

The nice trick: for each top-ranked paper, search GitHub for its official implementation — it's often linked straight from the arXiv abstract page, so you skip the guesswork entirely.

03

Video & other — the spoken explanation

Sometimes a 40-minute conference talk explains a method better than the paper. These sources fill the gaps the literature leaves.

SRC-06 · YouTube Data API

The lecture hall

The value isn't the video title — it's the words spoken inside it.

Search alone gives you titles and view counts, which are weak signals. What you actually want is the transcript, so the agent can read what was said and judge relevance. Transcript-focused MCP servers exist for exactly this.

Use it for: conference talks and university lectures that unpack a hard method.

Practical: rank by channel authority + views, and lean toward talks over hot-takes. Note YouTube's quota — a search costs ~100 units of a 10k/day default, so search sparingly and fetch transcripts for the keepers.

SRC-07 · Hugging Face

The ML workbench

For machine-learning topics, where the models and datasets actually sit.

Beyond a papers section, Hugging Face hosts model cards and dataset cards that describe how a method is used in practice — invaluable when the survey topic is an ML technique rather than pure theory.

Use it for: ML topics — pairing a paper with its runnable model and benchmark.

Practical: the Hub API is free; great as a bridge between "the paper" and "the thing you can load and run."

Bench note · the part that's easy to skip

Two streams will return the same paper (arXiv + Semantic Scholar + OpenAlex all carry it). De-duplicate on DOI or arXiv ID before writing, or the survey double-counts. And enforce one rule above all: the write-up may cite only what the tools actually returned — never a reference the model recalls from memory. Fabricated DOIs are the signature failure of research agents.

04

The pipeline — turning hits into a survey

Knowing the sources is half of it. The other half is the order you use them in — and here the order genuinely matters, because each step feeds the next.

  1. 1

    Scope the question

    Pin the topic, the time window, and what "related" actually means — ideally a similarity threshold, not a vibe. A narrow, well-defined question beats a broad one every time.

  2. 2

    Seed search

    One broad sweep (OpenAlex) for coverage and one sharp sweep (Semantic Scholar) for quality, to land 10–20 anchor papers you trust.

  3. 3

    Snowball the citations

    From each anchor, walk the citation graph both ways: backward through its references (the foundations) and forward through papers that cite it (the recent work). This is the step that turns a list into a survey.

  4. 4

    De-duplicate

    The same paper arrives from three sources. Collapse on DOI or arXiv ID, with a fuzzy title match as backup, before you count or rank anything.

  5. 5

    Rank & cluster

    Score every candidate, keep the top set, and group them into themes. Those clusters become the section headings of the survey — the structure writes itself.

  6. 6

    Synthesize & cite

    Write each theme in prose, citing only metadata the tools returned, and attach a BibTeX appendix. Nothing cited from memory.

Interactive · Step 3, the snowball

SEED foundations (older) recent work (newer)

Papers in the set: 1. Backward gives you the foundations; forward gives you what's new. Run both, or the survey skews to history or hype.

Fig. 3 — The snowball, live. One seed becomes a reading list by walking citations both directions.
05

Ranking & retrieval tactics

"Rank by relevance" is an instruction the agent can't follow until you say how. Two concrete tools: a scoring rubric, and query expansion.

Paper relevance

score = w₁·similarity + w₂·log(citations+1) + w₃·recency − w₄·off_topic

Citations reward old, famous work; recency rewards new work. Tune the weights so neither dominates.

Repository relevance

score = w₁·readme_match + w₂·log(stars+1) + w₃·commit_recency − w₄·staleness

Stars measure popularity, not fit — a stale 10k-star repo can be worse than a fresh 50-star one.

Video relevance

score = w₁·transcript_match + w₂·channel_authority + w₃·log(views+1)

Match on the transcript, not the title — that's where the actual content is.

Query expansion — your domain
ConceptTerms the agent should also try
Field-oriented controlFOC, vector control, dq-axis current control, PMSM current control
Sensorless positionback-EMF observer, sliding-mode observer, flux observer, sensorless FOC
NVH mitigationharmonic current injection, torque-ripple reduction, acoustic noise PMSM
Fault diagnosisfault detection & isolation, prognostics, bearing fault, inverter diagnostics

Feed these as OR-groups, and have the agent generate the synonyms from the topic — don't hand-maintain the list.

Interactive · the weights, live

MTPA control of PMSM
0
Sliding-mode sensorless observer
0
Survey of EV powertrains
0
HCI for acoustic-noise reduction
0
Generic deep-RL benchmark
0

Push citations up and the broad survey and the popular off-topic paper rise; push recency and the newest work wins; push off-topic and the deep-RL benchmark sinks. No single signal should run the table.

Fig. 4 — Five sample papers, re-ranked as you move the weights. This is "rank by relevance" made concrete.
06

Quotas & rate limits

The numbers you'll hit the moment you start building. Cache aggressively, back off on errors, and keep one source in reserve.

Rate limits — quick reference
SourceKeyRough limitNote
OpenAlexNo~100k/dayadd mailto for the polite pool
Semantic ScholarOptionallow → higher w/ keyget a free key for surveys
arXivNo~1 req / 3suse OAI-PMH for bulk
DBLPNobe gentleno hard cap — cache results
GitHubToken5,000/hrsearch capped lower (~30/min)
YouTube DataKey10,000 units/daya search ≈ 100 units
Hugging FaceOptionalgeneroustoken raises limits

Interactive · will this run blow the budget?

OpenAlex calls / day
GitHub searches / minute (burst)
YouTube units / day

YouTube is almost always the binding constraint — a single search costs ~100 units against a 10,000/day default. OpenAlex barely registers; GitHub's per-minute search cap bites if you fan out wide.

Fig. 5 — A back-of-envelope budget. Watch which meter turns red first as you scale the run up.
07

Failure modes — what goes wrong

Every one of these has bitten a research agent in the wild. Each card pairs a symptom with its fix.

Phantom citations

The model invents a plausible DOI or author list. Fix: cite only fetched metadata; check that every DOI resolves.

Echo duplicates

One paper counted three times across sources. Fix: dedup on DOI / arXiv ID before ranking.

Keyword blindness

GitHub or YouTube misses a perfect hit under a different name. Fix: query expansion plus a semantic re-rank.

Recency skew

Survey leans all-classics or all-last-month. Fix: balance backward and forward snowball; weight recency explicitly.

Paywall dead-end

Found the paper, can't read it. Fix: Unpaywall / CORE for a legal open copy; fall back to abstract + TLDR.

Silent truncation

Abstract-only when you needed full text, unflagged. Fix: have each tool report extraction success.

08

Further shelves — secondary sources

Useful additions once the core stack is working — mostly for resolving full text and mapping citation neighbourhoods.

CROSSREF
The DOI registry — canonical metadata and reference lists. Free. The source of truth for "is this DOI real?"
CORE
300M+ open-access full texts. When OpenAlex finds the paper, CORE often has the actual PDF.
OPENAIRE
European OA aggregator pulling from institutional repositories — another full-text route.
UNPAYWALL
Give it a DOI, get a legal free copy if one exists anywhere. The paywall escape hatch.
CONNECTED PAPERS
Visual citation neighbourhood around one seed paper — great for discovery, less for automation.
LITMAPS
Living citation maps that alert you when new related work appears.
GOOGLE SCHOLAR
Caution: no official API. Use it as a human discovery fallback — don't scrape it into the agent.
09

Person dossier — researching people, not topics

A different mode. Instead of a topic, the input is a name plus a disambiguator — an affiliation, a field, or an ID — and the output is a structured professional profile assembled from public sources.

NAME + disambiguator SCHOLARLY ORCID · OpenAlex · S2 PATENTS USPTO / PatentsView CODE & WEB GitHub · homepage DISAMBIGUATE Dossier confirmed profile LinkedIn — manual lookup only
Fig. 6 — A name in, a confirmed profile out. The gate is identity disambiguation; LinkedIn sits outside it, added by hand.
Where to look — and whether you can automate it
SourceWhat it givesAccess
ORCIDIdentity, affiliations, works, fundingOpen API
OpenAlexWorks, citations, institutions, co-authorsOpen API
Semantic ScholarPapers, h-index, citation graphOpen API
USPTO / PatentsViewPatents, co-inventors, assignees — name-disambiguatedOpen · sign-in from Jun 18
GitHubRepositories, activity, languagesOpen API
Web searchHomepage, talks, company bioOpen
Google Scholar— substitute with ORCID / OpenAlexNo API
LinkedInEmployment historyManual only
Disambiguation · the hard part

Many people share a name. Anchor each profile on a unique identifier — an ORCID iD, or a USPTO disambiguated inventor ID — then confirm every other hit by matching affiliation, field, and co-authors before you merge it in. Attributing a stranger's patents or papers to your subject is the people-search version of double-counting.

10

References

Primary documentation for every source above, plus ready-made multi-source MCP servers so you don't rebuild the retrieval layer.

  1. OpenAlex — API documentation. docs.openalex.org
  2. Semantic Scholar — Academic Graph API. semanticscholar.org/product/api
  3. arXiv — API user manual. info.arxiv.org/help/api
  4. DBLP — computer science bibliography. dblp.org
  5. IEEE Xplore — developer / API portal. developer.ieee.org
  6. GitHub MCP Server (official). github.com/github/github-mcp-server
  7. GitHub REST — search endpoints. docs.github.com/en/rest/search
  8. YouTube Data API v3. developers.google.com/youtube/v3
  9. Hugging Face — Hub API. huggingface.co/docs/hub/api
  10. paper-search-mcp — multi-source paper MCP (arXiv, S2, OpenAlex, dblp, …). github.com/openags/paper-search-mcp
  11. Academix — aggregated MCP w/ citation analysis & BibTeX. github.com/xingyulu23/Academix
  12. Crossref — REST metadata API. crossref.org · REST API
  13. CORE — open-access aggregator & API. core.ac.uk/services/api
  14. Unpaywall — DOI → free full text. unpaywall.org/products/api
  15. Connected Papers — citation neighbourhood explorer. connectedpapers.com
  16. ORCID — public API for researcher identity. info.orcid.org/documentation
  17. USPTO Open Data Portal (PatentsView). data.uspto.gov