Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
Parijat Chatterjee avatar

Research Compiler Local

  • Updated May 25, 2026
  • Parijat-18/research-compiler

research-compiler-local is a Claude Code marketplace plugin that compiles an arXiv paper's citation neighborhood into a Graph RAG sqlite store, research brief, and llm-wiki for developers reproducing ML research papers.

About

research-compiler-local distributes the paper-compiler plugin that turns one arXiv ID into a queryable research corpus beside your code. A background CLI resolves references, acquires PDFs from arxiv, Semantic Scholar, OpenAlex, and Unpaywall, parses chunks, extracts implementation atoms, detects communities, and builds research.db with sqlite-vec and FTS5. Verified JEPA demo numbers cite roughly 434 neighborhood papers, 165 deduped atoms, 19,500 indexed chunks, and a ~72 MB database served by 26 read-only MCP tools. Skills include /paper-compiler:build-research-context, auto-invoked use-research-context and audit-against-research flows, plus wiki-query, wiki-ingest, and wiki-lint commands. The compile CLI never serves runtime queries; the MCP server never writes—separation keeps stale DB diagnosis independent from tool fixes. Documented five-turn sessions consume about 25K MCP-result tokens versus roughly 6× for repeated web search. Install from the local marketplace scaffold at ./plugin-scaffold when developing the plugin itself.

  • Compiles one arXiv ID into ~434 papers and 19,500 indexed chunks
  • Exposes 26 read-only MCP tools over sqlite-vec plus FTS5 research.db
  • Produces ≤8000-token research.md brief and Karpathy-style llm-wiki tree
  • 14 implement-* sub-skills whitelist 4–6 MCP tools per category
  • Verified build extracts 165 implementation atoms across 9 categories

Research Compiler Local by the numbers

  • Data as of Jul 7, 2026 (Skillselion catalog sync)
/plugin marketplace add Parijat-18/research-compiler

Add your badge

Show developers this marketplace is listed on Skillselion. Paste this into your README.

Listed on Skillselion
Last updatedMay 25, 2026
RepositoryParijat-18/research-compiler

How do you compile paper citations into Graph RAG for Claude?

Run /paper-compiler:build-research-context on an arXiv ID to generate research.md, research.db, and 26 MCP tools for grounded paper implementation in Claude Code.

Who is it for?

ML engineers using Claude Code to reproduce papers who need citation-neighborhood context, atom-level audits, and wiki-promoted cross-paper answers in-repo.

Skip if: Quick literature summaries without implementation intent, repos unwilling to host a ~72 MB sqlite corpus, or teams without Claude Code MCP support.

What you get

research.md brief, research.db Graph RAG store, evidence atom markdown files, wiki articles, and read-only MCP tool access.

  • research.db
  • research.md brief
  • wiki/ knowledge base

By the numbers

  • Exposes 26 read-only MCP tools over research.db
  • Verified JEPA build indexes 19,500 chunks and 165 implementation atoms
  • Compiles neighborhoods of roughly 434 papers from one arXiv seed

Recommended Marketplaces

How it compares

Pick research-compiler over raw PDF uploads when reproduction needs citation-neighborhood Graph RAG, atom audits, and wiki-promoted cross-paper answers inside the repo.

FAQ

How many MCP tools does research-compiler expose?

research-compiler exposes 26 read-only MCP tools through the paper-compiler MCP server, up from 15 in earlier versions. Tools include snippet-first query_chunks and paper_text previews plus dependency tracing for implementation atoms.

What artifacts land in the research/ directory?

research-compiler writes research.md (≤8000-token brief), research.db Graph RAG store, SCHEMA.md, per-atom evidence markdown under evidence/, and a regenerable wiki/ tree with atoms, papers, communities, and promoted answers.

Does the MCP server modify research.db during coding?

research-compiler keeps compile-time writes in the CLI only; the MCP server is read-only at runtime. Re-ingest or rebuild via /paper-compiler:wiki-ingest or a fresh compile when the corpus must change.

AI Agentsresearchautomation

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.