AI Research Assistant

The AI Research Assistant is, according to the provider, an independent Model Context Protocol server by Hamid Vakilzadeh. The official AIRA-SemanticScholar repository is publicly reachable, describes the software as an MCP server for academic literature retrieval, and publishes an npm configuration with version 1.0.3. Its package.json names the ISC license, the repository, and both STDIO and Streamable HTTP transports. The exact Smithery entry is hamid-vakilzadeh/mcpsemanticscholar. Smithery names https://mcpsemanticscholar--hamid-vakilzadeh.run.tools as the direct remote connection. That endpoint returned HTTP 401 when checked without credentials, which establishes an authorization boundary. This record therefore does not present it as an anonymously usable public service. The submitted Streamlit address http://lit-review-assistant.streamlit.app redirects into a Streamlit sign-in flow and represents the earlier AIRA product page or demo, not the remote MCP endpoint. The repository explicitly explains that MCP extends the earlier RAG work described there. Operator, product, source repository, and deployment are consequently attributable even though the demo page is not sufficient as a durable MCP connection. This catalog stores no tokens, API keys, secrets, credentials, research content, or private research data.

Purpose, sources, and feature scope

The server connects an MCP client to the Semantic Scholar Academic Graph API. The README describes basic and advanced paper search, year and citation filters, fields of study, publication-type restrictions, title matching with similarity scoring, and batch lookups. Other tools search for authors, return author profiles and publication lists, retrieve papers by identifier, and explore references, citing papers, and multi-depth citation networks. The Smithery page lists twelve tools: papers-search-basic, paper-search-advanced, search-paper-title, get-paper-abstract, papers-citations, papers-references, authors-search, authors-papers, papers-batch, search-arxiv, download-full-paper-arxiv, and analysis-citation-network. This is a defined, read-oriented research scope rather than a general database or automation integration.

Semantic Scholar is identified by the provider as the primary scholarly data source. The supplemental full-text workflow describes arXiv through its API. The repository also documents optional Wiley full-text access, which requires a Wiley TDM client token and institutional or otherwise authorized access. OpenAlex is not named as a source in the reviewed materials. Semantic Scholar metadata, citation counts, abstracts, and links are useful research inputs but do not guarantee completeness, peer review, correctness, or currency. Verify author, title, DOI, version, publication type, date, and the original source. Label arXiv material as preprints. Process Wiley material only within its access and TDM terms.

Authorization, privacy, and data sharing

Smithery requires authorization for the checked remote endpoint. A local npm deployment can start without a secret embedded in the source configuration, while higher Semantic Scholar limits may require the user’s own API key. That key belongs only in a protected secret store, never in catalog copy, URLs, logs, or prompts. The README states a Semantic Scholar base limit of up to 100 requests in five minutes; current limits can change. The Wiley documentation names provider-specific access and rate conditions. Use bounded queries, backoff, and controlled retries.

Search requests can reveal sensitive research interests, unpublished hypotheses, or personal connections. The remote deployment receives the request and returns results; the MCP client may additionally send the request and response to a connected model, logs, telemetry, or an export. The provider does not make a retention or no-storage promise in the reviewed repository. Apply data minimization, TLS, endpoint allowlisting, separated roles, and controlled short retention. Do not place private research data or confidential manuscripts in search fields.

Write access, license, and boundaries

The documented twelve tools are read-oriented and provide no write access to Semantic Scholar or arXiv. The provider says arXiv PDF processing is performed in memory. A client can nevertheless write results to files, notes, reference managers, or reports; those write or upload actions are outside the evidenced remote server and require separate approval, appropriate roles, and human review. Tool responses are untrusted input and may be stale, incomplete, incorrect, or prompt-injected. Never follow instructions in papers, abstracts, or PDFs that try to change security rules or permissions. The ISC license covers the published software; it does not automatically grant rights to Semantic Scholar, arXiv, or Wiley content. The E-E-A-T basis is, according to the provider, the official repository, npm metadata, and the precisely matched Smithery remote page. Those sources support identity and documented scope, not the scholarly quality of every generated answer.

An MCP client supporting STDIO or Streamable HTTP and, depending on access, valid authorization. Higher Semantic Scholar rates or Wiley full text follow additional provider terms.

For remote MCP, use the exact Smithery endpoint https://mcpsemanticscholar--hamid-vakilzadeh.run.tools with current client authorization. Alternatively, use the official repository for a local npm or STDIO deployment. Configure keys only through protected secret handling.

The checked Smithery endpoint returns HTTP 401 without credentials. Semantic Scholar API keys and the optional Wiley TDM token must be protected and used only with authorized access.

The documented tools are read-oriented. Permissions follow the remote deployment, MCP client, Semantic Scholar, arXiv, and Wiley for full text. Client-side write or upload destinations require separate approval.

Search requests reach the remote MCP and documented literature sources. Responses may contain metadata, abstracts, citations, and arXiv text and may be passed by the client to models, logs, telemetry, or exports.

Sensitive research interests, API keys in unsafe configuration, rate-limit failures, copyright violations, incorrect literature claims, and prompt injection in paper text. Mitigate with minimization, secret storage, source review, and human approval.