opencode-research-papers

An opencode plugin that adds a research_papers tool. This is an AI-facing tool, not a slash command — ask the AI to search for papers and it will call the tool for you.

Install. Restart. Ask for papers. It works.

Features

• No API keys required. Uses arXiv for fresh preprints and OpenAlex for broader scholarly metadata, citation counts, and open-access links.
• Smart auto source routing — arXiv for latest, OpenAlex for top_cited and trending, both merged when available.
• Markdown output with title, authors, date, PDF link, abstract, and citation count where available.
• Filter by recency (latest, trending) or citation impact (top_cited).
• Narrow results to the past week, month, or year.
• strict mode applies anchor + concept-group filtering to reduce loosely matched results.
• Source-level fallback: if one source fails or is rate-limited, the other handles the request transparently.
• arXiv requests are spaced 3 seconds apart per arXiv's public API terms.

Installation

Add this to your opencode.json:

{
  "plugin": [
    "opencode-research-papers"
  ]
}

Restart opencode. The tool registers automatically.

Updating

OpenCode caches plugin packages and does not auto-update on restart. After updating to a new version, clear the cache before restarting:

Windows (PowerShell):

Remove-Item -Recurse -Force "$env:USERPROFILE\.cache\opencode\packages\opencode-research-papers@latest"

macOS / Linux:

rm -rf ~/.cache/opencode/packages/opencode-research-papers@latest

Then restart opencode.

Usage

This is an AI tool — you do not invoke it with a slash command. Ask naturally:

"Find the latest papers on Image Segmentation"

"Show me trending Scene Text Recognition papers"

"Get top-cited Retinal Vessel Segmentation papers"

"Find 20 latest papers on Generative Adversarial Networks from the last month"

Configuration

Pass options as a tuple:

["opencode-research-papers", {
  "defaultMaxResults": 15,
  "defaultSource": "auto"
}]

| Option | Default | Description | |--------|---------|-------------| | defaultMaxResults | 10 | How many results to return (1–50) | | defaultSource | "auto" | Source routing: auto, arxiv, or openalex |

Source routing

The auto default selects the primary source per filter:

| Filter | Primary | Fallback | |--------|---------|----------| | latest | arXiv | OpenAlex | | top_cited | OpenAlex | arXiv | | trending | OpenAlex | arXiv |

Override with source: "arxiv" or source: "openalex" to use a single source.

Data Sources

arXiv

Provides fresh preprints across AI, CS, math, and physics. Delivers direct PDF links and clean metadata. No signup required. Requests are rate-limited to one per 3 seconds per arXiv's public API terms.

OpenAlex

Broader scholarly search with citation counts, DOI metadata, and open-access links. Works without an API key at 10 requests per second.

Error Handling

If one source is down or rate-limited, the plugin returns the other source's results with the specific error noted. If both fail, you receive a consolidated error message. The tool does not crash.

License

MIT

Roadmap

Potential future additions:

• GitHub paper-list repos — Search curated repository lists (e.g. scene-text-detection-recognition-papers) alongside paper results.
• Better deduplication — Title normalization is a rough heuristic; smarter deduplication could avoid showing the same paper twice.
• Synonym expansion — Expand common terms in strict mode (e.g. GAN → cGAN, WGAN, StyleGAN) to improve recall.
• Semantic strict mode — Use embeddings or cross-encoder reranking instead of keyword matching for relevance filtering.
• Citation counts for arXiv papers — Cross-reference arXiv results with OpenAlex to retrieve citation data.