pwc CLI
pwc-cli
Hugging Face's CLI, skill and MCP server: an agent searches papers, benchmarks and top models in Papers with Code with no account or key
- 217 ★ on the repository
- Last commit
Extended access
Why this level
- The installer downloads a binary from GitHub and places it in `~/.local/bin` after a SHA-256 check
- Requests go to `paperswithcode.co`, and `pwc paper edit submit` publishes paper edits immediately under your Hugging Face name
Install
- Works in
Manual install
curl -LsSf https://raw.githubusercontent.com/huggingface/pwc-cli/main/standalone_cli/install.py | python3The installer downloads the binary, verifies SHA-256 and installs it to `~/.local/bin/pwc`. Python 3 is needed for the installer itself.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The installer downloads a binary from GitHub and places it in `~/.local/bin` after a SHA-256 check
- Requests go to `paperswithcode.co`, and `pwc paper edit submit` publishes paper edits immediately under your Hugging Face name
This is third-party code. Review the repository files before installing.
What it does
pwc is a command-line client for the Papers with Code catalog, which Hugging Face runs. An agent or a person searches papers by title, topic, author and arXiv ID, reads metadata and the full text of a work, and looks at predecessors, successors, recent and trending publications. Separate commands show tasks, methods, conferences, organizations, frameworks and benchmarks with top models, scores and parameter counts. Public queries are anonymous and read-only: no account, key or Python environment is needed. In a terminal the result is a table, when piped it is TSV, and the `--json` flag returns JSON with a schema version. Exit codes are stable: 0 means success, 2 an argument error, 3 a network or server failure, 4 an invalid API response. The same repository holds a skill and an MCP server. `pwc skills add` puts a skill matched to the installed CLI version into the project. The hosted server at `https://paperswithcode.co/mcp` exposes one tool per research command. Editing a single confirmed paper has its own flow with a browser sign-in.
Who it is for. Machine learning researchers and engineers who hand literature review to an agent: find papers on a topic, compare models on a benchmark, check whether official code exists.
Good fit when
- You need an overview of recent work on a topic: `pwc paper recent` and `pwc paper trending --max-age-days 90` give lists, and `pwc paper related` finds neighboring papers
- An agent picks a model for a task and checks the best results: `pwc benchmark --name wikitext-103 --max-parameters 3B` limits output by model size
- Before reading a paper it helps to know whether official code exists: the `--has-official-implementation` flag drops works without a linked repository
- You are tracing a method's history and want what came before and after a work: `pwc paper lineage list` shows predecessors and successors
- A paper has to be discussed with the agent in full: `pwc paper read` prints the stored text of the work
Not a fit when
- The work is not in the Papers with Code catalog: search covers only its database, and the CLI does not query PubMed or other sources
- You need an offline mode: every command calls the API at `https://paperswithcode.co/api/v1`
- You are on Windows: prebuilt binaries exist for Linux x86-64, macOS Intel and Apple Silicon, so Windows means installing from source with uv or pipx
- You want to edit the catalog in bulk: the CLI changes one confirmed paper and has no ingestion, image or embedding commands
Example request
Find five recent small vision-language model papers in Papers with Code and tell me which ones have official codeLimitations
- Editing works only on a server with the paper-edit API enabled: a browser sign-in link opens write access to one confirmed paper for one hour.
- A reply of "no official code" only means the catalog has no linked repository.
- Prebuilt binaries exist for Linux x86-64, macOS Intel and Apple Silicon; other systems need Python 3.10 or newer.
- GitHub does not detect a license for the repository, so check the usage terms with the authors.
How to disable. The binary sits at `~/.local/bin/pwc` and can simply be deleted. The skill is removed by deleting the `pwc` folder from `.agents/skills` (or `.claude/skills`), and the MCP server is removed with `claude mcp remove paperswithcode` or by editing the client configuration.
FAQ
How do I install the pwc skill in Claude Code?
Install the CLI and run `pwc skills add --claude` for a project or `pwc skills add --claude --global` for all projects. The CLI generates the skill itself, so it matches the CLI version. An existing skill is left alone; add `--force` to replace it.
How do I connect the Papers with Code MCP server to Claude Code?
Run `claude mcp add --transport http paperswithcode https://paperswithcode.co/mcp`. No sign-in or key is needed. For Codex the command is `codex mcp add paperswithcode --url https://paperswithcode.co/mcp`.
Do I need an account to search papers?
No, search, paper reading and benchmarks are anonymous. Sign-in is needed only to edit a paper: `pwc auth login --paper 123` opens a browser link.
How do I get output as JSON?
Add `--json` to any data command, for example `pwc search "language models" --limit 5 --json`. The response carries `schema_version` and `data` fields.
How do I refer to a paper in commands?
The `PAPER` argument takes an arXiv ID, a numeric external-paper ID or an exact title in quotes. An ambiguous title is rejected and the matching IDs are printed.
Discussion
No questions or reviews yet.
Related
CLI toolsAn MCP server that splits math papers from arXiv, LaTeX and PDF into theorems, proofs and dependencies backed by the source
A universal MCP CLI client from Apify: persistent sessions, OAuth, JSON output for scripts and shell access to any MCP server for agents
A CLI agent that works with project files and shell commands, lets you choose a model provider and connects external MCP tools.
A zero-dependency Python CLI: one catalog of MCP servers and skills for every agent, with only a compact index in context
Remote MCP server that tells which search, extraction, reranker or retrieval API fits your task best, with confidence intervals.
A CLI that turns any MCP server, OpenAPI spec or GraphQL endpoint into commands at runtime, with no codegen
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| pwc CLI (this card) | yes | 217 | ||
| arxiv-mcp-server | no | not required | 3,273 | a local arXiv server: reading papers by section and BibTeX from metadata |
| Paper Search MCP | no | key is optional | 3,062 | searches and downloads papers from arXiv, PubMed, bioRxiv and other open sources |
README in short
The page opens with the hosted MCP server section and connection commands for Claude Code, Codex, Cursor and VS Code, then covers binary and source installs, adding the skill and a quick start with search and filter examples. A command table, output and exit-code rules, the `PWC_API_URL` variable for another API and a scope section follow: public reads are anonymous, editing is limited to one paper.