# web-researcher-mcp

> An open-source MCP server that gives AI assistants real web search with verified citations, full-article reading, and trusted source filtering via search lenses.

web-researcher-mcp is an open-source Model Context Protocol (MCP) server written in Go that equips AI assistants like Claude, Cursor, and any MCP-compatible client with honest, citation-verified web research capabilities. Built by Zohar Babin and released under the MIT license, it runs entirely on the user's machine — queries never pass through a third-party server. The latest release is v1.34.0, published June 16, 2026, which added a Python SDK.

## What It Is

web-researcher-mcp is an MCP server that plugs into AI assistants and gives them a suite of research tools: web search, full-page scraping, academic paper lookup, patent search, SEC filing retrieval, US court record search, citation verification, and more. Unlike AI search products that synthesize results from opaque indexes, this tool fetches and reads primary sources directly, returning working links and properly formatted citations (APA, MLA, BibTeX, RIS, CSL-JSON). The core design goal is eliminating hallucinated citations — every result must be a real, verifiable URL.

## How Search Lenses Work

A central feature is "search lenses" — curated domain lists that restrict which websites an AI is allowed to search. Instead of querying the open web and surfacing SEO spam or AI-generated content, a lens limits results to trusted sources for a given field:

- **`medical`** — PubMed, WHO, NIH, and clinical sources
- **`legal`** — law databases, case repositories, statutes
- **`academic`** — preprint servers, open-access journals, repositories
- **`security`** — CVEs, advisories, vulnerability research
- **`finance`** — SEC EDGAR, market filings
- **`journalism`** — public records, FOIA, corporate filings

Users can also define custom lenses by listing trusted domains in a JSON file. Lenses work with every supported search provider.

## Tool Suite and Research Capabilities

The server exposes over 20 tools to connected AI assistants, covering the full research workflow:

- **`web_search`** / **`search_and_scrape`** — web search with optional lens filtering and quality scoring
- **`scrape_page`** — full-text extraction of web pages, PDFs, Word docs, YouTube transcripts, and Hacker News threads
- **`academic_search`** — real papers with DOIs, authors, citation counts, and open-access links via OpenAlex/CrossRef
- **`citation_graph`** — walks a paper's citation neighborhood with intent/influence signals
- **`patent_search`** — US, European, and international patent offices
- **`filing_search`** — SEC EDGAR filings (10-K, 10-Q, 8-K) and structured XBRL company facts
- **`legal_search`** — US court opinions and dockets via CourtListener
- **`verify_citation`** / **`audit_bibliography`** — checks whether citations exist, match real records, and are not retracted
- **`answer`** — synthesized factual answers with citations (requires Exa provider)
- **`sequential_search`** — multi-step deep research with session persistence
- **`research_export`** — exports sessions as markdown or JSON reports with full provenance

Guided prompt templates (`comprehensive-research`, `fact-check`, `competitive-analysis`, `literature-review`) are also bundled and surface as slash commands in supported AI apps.

## Deployment and Setup Paths

The server ships as a single signed binary with multiple install paths:

- **Python/uvx** — `uvx web-researcher-mcp` (no Go required, cross-platform)
- **Homebrew** — `brew install zoharbabin/tap/web-researcher-mcp` (macOS/Linux)
- **One-command shell installer** — curl/PowerShell scripts for macOS, Linux, and Windows
- **WinGet / Scoop** — Windows package managers
- **Docker** — `zoharbabin/web-researcher-mcp` image with Chromium bundled
- **Go install** — `go install` from source
- **One-click buttons** — deep-link installers for Cursor, VS Code, and LM Studio

DuckDuckGo is the zero-config default (no API key needed). Optional providers — Google PSE, Brave, Serper, Tavily, Exa, SearXNG, SearchAPI.io, and Hacker News — can be added by setting environment variables. Multiple providers can be configured for automatic failover via `SEARCH_ROUTING`.

## Update: v1.34.0 — Python SDK

The v1.34.0 release (June 16, 2026) introduced a Python SDK (`web-researcher-mcp` on PyPI) with an async `WebResearcherClient` and a sync wrapper for scripts and notebooks. The SDK is installable via `pip`, `uv`, or `uvx`, and the PyPI wheels vendor the compiled Go binary so no Go toolchain is required. The project also ships an HTTP server mode with OAuth 2.1 support for team deployments, and the repository shows active development with 27 stars and 4 forks as of the latest data.

## Features
- Web search with configurable search providers (DuckDuckGo, Google PSE, Brave, Serper, Tavily, Exa, SearXNG, SearchAPI.io, Hacker News)
- Search lenses for restricting results to trusted domain lists by field
- Full-page scraping of web pages, PDFs, Word docs, YouTube transcripts, and Hacker News threads
- Academic paper search with real DOIs, authors, citation counts, and open-access links
- Citation graph traversal with intent and influence signals
- Patent search across US, European, and international patent offices
- SEC EDGAR filing search and structured XBRL company facts
- US court opinion and docket search via CourtListener
- Economic data lookup (World Bank, OECD, Eurostat, FRED)
- ClinicalTrials.gov search for clinical trial registrations
- Citation verification and bibliography auditing
- AI recommendation list auditing for conflicts of interest and dead links
- Internet Archive snapshot capture via Save Page Now
- Synthesized factual answers with citations (requires Exa)
- Structured JSON extraction from search results
- Multi-step sequential deep research with session persistence
- Research session export as markdown or JSON with full provenance
- Bibliography formatting in APA, MLA, BibTeX, RIS, and CSL-JSON
- Guided prompt templates for comprehensive research, fact-checking, competitive analysis, and literature review
- Multi-provider automatic failover routing
- HTTP server mode with OAuth 2.1 for team deployments
- Python SDK with async and sync wrappers
- SSRF protection, rate limiting, and content sanitization
- Disk-based result caching
- JavaScript-heavy site rendering via headless Chromium

## Integrations
Claude, Claude Desktop, Claude Code, Cursor, VS Code, LM Studio, DuckDuckGo, Google Programmable Search Engine, Brave Search, Serper, SearchAPI.io, SearXNG, Tavily, Exa, Hacker News Algolia, OpenAlex, CrossRef, Semantic Scholar, SEC EDGAR, CourtListener, ClinicalTrials.gov, World Bank, OECD, Eurostat, FRED, EPO, USPTO, The Lens, Internet Archive Wayback Machine, Docker, Homebrew, PyPI, WinGet, Scoop

## Platforms
WINDOWS, MACOS, LINUX, WEB, API, VSC_EXTENSION, DEVELOPER_SDK, CLI

## Pricing
Open Source

## Version
v1.34.0

## Links
- Website: https://zoharbabin.github.io/web-researcher-mcp/
- Documentation: https://zoharbabin.github.io/web-researcher-mcp/
- Repository: https://github.com/zoharbabin/web-researcher-mcp
- EveryDev.ai: https://www.everydev.ai/tools/web-researcher-mcp