Quickstart

Scholar is a local MCP server that runs on the same computer as your agent. This page uses Claude Code as the example. Other clients follow the same steps, and Install in your client has their configuration.

This guide installs Scholar MCP 0.9.0 from PyPI, including the CLI and scientific data tools. Plugins, containers and desktop bundles use the same release. The compatibility wheel index supplies native dependencies for Intel macOS and Windows on ARM.

1. Install uv

   Scholar runs through uv, which prepares a separate Python 3.12 for it. If uv --version already works, skip this step.

   macOS and Linux

   curl -LsSf https://astral.sh/uv/install.sh | sh

   Windows

   powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

2. Connect your agent

   Scholar CLI

   Install the scholar-mcp command, then let it write the client configuration.

   uv tool install --python 3.12 \
     --find-links https://github.com/Liyux3/scholar-mcp/releases/download/v0.9.0/wheel-index.html \
     "scholar-mcp[rerank]==0.9.0"
   scholar-mcp install claude-code

   install backs up a file before changing it and can be run again safely. --all configures every client it detects, --dry-run only shows the change, and --profile research adds the citation graph and paper library. scholar-mcp update upgrades the installed command. For a client pinned to a uvx version, rerun the installer with the new release.

   Manual

   To skip installing the command, register the server with uvx. On first launch uvx prepares Python 3.12 and the local ranking model.

   claude mcp add scholar -- uvx --python 3.12 \
     --find-links https://github.com/Liyux3/scholar-mcp/releases/download/v0.9.0/wheel-index.html \
     --from 'scholar-mcp[rerank]==0.9.0' scholar-mcp

   This registers the six core tools for the current project. --scope user makes it apply to every project, and -e SCHOLAR_MCP_EXTENSIONS=research before -- adds the citation graph and paper library.

   Ask your agent

   Works in any client, including ones not listed anywhere on this site. Paste this into your agent:

   Install Scholar MCP into the client you are running in, following
   https://raw.githubusercontent.com/Liyux3/scholar-mcp/v0.9.0/llms-install.md
   Then tell me to restart, and check that search_papers is available.

   The guide tells the agent which command to start and how common clients write it down, so the agent works out its own client’s format.

3. Check the connection

   Start a new Claude Code session and run /mcp. scholar should show as connected. In a terminal, scholar-mcp doctor shows the version, the registered clients, and which sources and ranking model are available.

4. Ask a research question

   Ask the way you would ask a colleague. The agent chooses the tools itself.

   Find recent work on evaluating long-term memory in LLM agents,
   read the most cited one and compare its benchmarks.

   The agent searches with search_papers, opens the chosen paper with read_paper and answers from the text. Follow up with who cites this? or save these three to the memory collection.

5. Add API keys when needed

   Every key is optional. Keys widen coverage and raise limits. Save each one once with scholar-mcp keys set NAME and every client uses it, or put them in the env block of one client’s configuration.

   Server environment

   "env": {
     "S2_API_KEY": "your-semantic-scholar-key",
     "OPENALEX_API_KEY": "your-openalex-key",
     "DASHSCOPE_API_KEY": "your-dashscope-key"
   }

   If you add only two, make them the free ones that lift the tightest limits: an OpenAlex key (without one OpenAlex allows about a hundred searches a day) and a registered OpenAIRE service, OPENAIRE_CLIENT_ID and OPENAIRE_CLIENT_SECRET (60 requests an hour without it, 7,200 with it). A Semantic Scholar key raises its limits and enables snippet search and recommendations. An OpenAlex key is used for its keyword search, semantic search and citation data. A DashScope key moves ranking to the cloud Qwen3 reranker. All variables are in Configuration.

Research profile

The core profile has six tools. Literature reviews usually also want build_paper_graph and paper_library. Add --profile research when installing, or set SCHOLAR_MCP_EXTENSIONS=research. The Claude Code and Codex plugins enable it by default and include the Deep Research skill.

Next steps

Research workflowsSearch, screen, read and save, quickly and well.

Command lineEvery tool in the terminal, plus install, doctor and update.