MCP Paperswithcode

What hbg/mcp-paperswithcode MCP server does

The hbg/mcp-paperswithcode MCP server gives an MCP-compatible assistant access to research information from PapersWithCode. Its coverage includes papers, authors, research areas, conferences, conference proceedings, benchmark results, tasks, methods, datasets, and repositories linked to papers.

Paper discovery supports optional title, abstract, and arXiv ID filters. After finding a paper, an agent can retrieve its metadata and inspect related items such as benchmark results, tasks, methods, datasets, and code repositories. The server also accepts a direct PDF or HTML URL and returns extracted readable paper text.

How it works

The server acts as an MCP client for the PapersWithCode API. An assistant calls the exposed tools with identifiers or search values, and the server returns the corresponding API information. Most detail-oriented operations use IDs returned by an earlier search or lookup, such as a paper ID, author ID, area ID, conference ID, or proceeding ID.

A typical workflow is to search for papers by title, abstract text, or arXiv ID, retrieve a selected paper, and then query its associated tasks, methods, datasets, results, or repositories. Author workflows can search by full name and list papers by either the selected author’s ID or name. Conference workflows move from a conference to its proceedings and then to papers within a proceeding.

Tools and capabilities

The hbg/mcp-paperswithcode MCP server groups its tools by research object:

• Search and retrieve papers, including metadata and readable content from a supplied URL.
• List a paper’s benchmark results, tasks, methods, datasets, and linked repositories.
• Search research areas by partial or full name and retrieve area metadata or related tasks.
• Search authors by full name, retrieve author metadata, and list papers associated with an author.
• List conferences with an optional name filter, retrieve conference metadata, and inspect proceedings.
• Retrieve a specific proceeding and list papers belonging to it.

Paper searches accept optional filters rather than requiring every field. Other operations require the relevant identifier, while the paper-reading tool requires a direct paper URL.

Setup and configuration

The README documents installation through Smithery for Claude Desktop using the Smithery command-line tool. It does not provide a separate project-specific package command, hosted endpoint, or local configuration example in the supplied material. The documented setup path is therefore tied to Smithery and Claude Desktop rather than a confirmed standalone installation command for the repository itself.

No API key, token, or environment variable is specified in the supplied documentation. The README also does not describe configurable request settings, storage, authentication options, or response caching.

Limitations and notes

Search and lookup behavior depends on the information available through the PapersWithCode API. Paper-content extraction requires the caller to provide a direct PDF or HTML URL; the supplied material does not state that the server discovers or downloads arbitrary papers automatically.

Several tools require identifiers. An agent may need to perform a search first to obtain the correct paper, author, research-area, conference, proceeding, or other ID. The documentation lists supported tool inputs but does not specify API rate limits, pagination behavior, error handling, freshness guarantees, or coverage beyond the listed PapersWithCode resources.