The combined search is the reason to attach this rather than five separate servers: at the start of a problem you rarely know whether the answer is a Stack Overflow thread, an MDN page or a package that already does it, and running all five in parallel costs one round trip. The per-platform tools are there for when you do know. Anonymous GitHub search will rate-limit under sustained use, so add a token before relying on it.
An MCP server that puts the five places a developer actually looks behind one set of tools, so a model researching a problem can reach answers, docs, code and packages without a browser. Results are cached for an hour, so repeated queries during one session do not re-hit the upstream APIs.
- Stack Overflow questions with scores, answer counts and excerpts — `search_stackoverflow`
- MDN Web Docs matches with summaries and links, for the web-platform questions where the spec is the answer — `search_mdn`
- GitHub searched in two passes at once: top repositories by stars, and matching code files with their repository context, optionally filtered to one language — `search_github`
- Npm packages with version, description and download stats, and PyPI packages with version, author and links — `search_npm`, `search_pypi`
- All five platforms searched in parallel and returned as one combined result — `search_all`
No account needed to start: every platform is queried anonymously. A GitHub personal access token in `GITHUB_TOKEN` is optional and raises the rate limit on the GitHub half. Node.js >= 20.11.0 and npm >= 10.0.0. The npm package name is `code-research-server` (0.1.0) and it is marked private, so install from a clone with `npm install` and `npm run build`, or let the Smithery installer write the client entry. Licensed AGPLv3.
One command — npx -y @smithery/cli install @nahmanmate/code-research-mcp-server --client claude