claude mcp add semantic-scholar -s user -- uvx --from git+https://github.com/akapet00/semantic-scholar-mcp semantic-scholar-mcp
Claude Desktop
Add to your config file (~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%\Claude\claude_desktop_config.json on Windows):
claude mcp add semantic-scholar -s user -e SEMANTIC_SCHOLAR_API_KEY=your_key -- uvx --from git+https://github.com/akapet00/semantic-scholar-mcp semantic-scholar-mcp
Claude Desktop / Cursor / Windsurf / OpenCode — add an "env" block to the config:
"env": {
"SEMANTIC_SCHOLAR_API_KEY": "your_key"
}
SSL Issues
If you're behind a corporate proxy or firewall and see CERTIFICATE_VERIFY_FAILED errors, set DISABLE_SSL_VERIFY=true as an environment variable (same approach as the API key above). Only use this in trusted networks.
Why MCP?
The same prompt was run with and without this MCP server connected to Claude Code (Opus 4.6):
Find the first journal paper by Ante Kapetanovic and extract all references from that paper.
Without MCP
With MCP
Recall
85.3% (58/68 refs)
100% (68/68 refs)
Tool calls
40 (75% failed)
14 (43% failed)
Agent turns
30
10
Side effects
pip install, temp files
None
Without MCP, the agent spent most of its turns fighting HTTP errors, parsing raw HTML, and installing packages. With MCP, the author and paper were found in 2 turns. The same publisher restriction on references was hit, but recovery was faster and every reference was retrieved via a clean fallback.
When installed via git+https://..., updates are not pulled automatically. To get the latest version:
# Refresh the cached package
uvx --refresh --from git+https://github.com/akapet00/semantic-scholar-mcp semantic-scholar-mcp
# Or re-add in Claude Code
claude mcp remove semantic-scholar -s user
claude mcp add semantic-scholar -s user -- uvx --from git+https://github.com/akapet00/semantic-scholar-mcp semantic-scholar-mcp
You can also pin to a specific branch, tag, or commit:
git clone https://github.com/akapet00/semantic-scholar-mcp.git
cd semantic-scholar-mcp
uv sync
# Run tests
uv run pytest
# Run integration tests (requires network access)
uv run pytest -m integration
# Lint and format
uv run ruff check src/ tests/
uv run ruff format src/ tests/
# Type checking
uv run ty check src/
# Dev mode with inspector
uv run fastmcp dev src/semantic_scholar_mcp/server.py
Configuration
See CONFIGURATION.md for all environment variables (cache, retry, circuit breaker, logging, default limits).