Multi-source web research via the web-researcher-mcp Go server (deployed binary v1.43.0) — you pick the search ENGINE and the trusted SOURCES, and every citation is a real, checkable link. 21 live MCP tools (verified against tools/list): web/image/news/academic/patent search, search_and_scrape, sequential_search, awesome_list_search, brand_research; domain search (clinical_search, legal_search/CourtListener, econ_search/World Bank+FRED); scrape_page (full PDF/DOCX/PPTX/YouTube/HN, not snippets); citation integrity (verify_citation, verify_recommendation, audit_bibliography, citation_graph, archive_source/Wayback, format_bibliography APA/MLA/BibTeX/RIS/CSL); research session memory + export (get_research_session, research_export). Search LENSES restrict results to trusted domains (academic, clinical, legal, finance, government, journalism, devops, docs). Pluggable backends (Google PSE/Brave/Serper/SearXNG/SearchAPI/Exa), DuckDuckGo fallback. Use for reputation-attached research needing verifiable citations — t
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Multi-source web research via the web-researcher-mcp Go server (deployed binary v1.43.0) — you pick the search ENGINE and the trusted SOURCES, and every citation is a real, checkable link. 21 live MCP tools (verified against tools/list): web/image/news/academic/patent search, search_and_scrape, sequential_search, awesome_list_search, brand_research; domain search (clinical_search, legal_search/CourtListener, econ_search/World Bank+FRED); scrape_page (full PDF/DOCX/PPTX/YouTube/HN, not snippets); citation integrity (verify_citation, verify_recommendation, audit_bibliography, citation_graph, archive_source/Wayback, format_bibliography APA/MLA/BibTeX/RIS/CSL); research session memory + export (get_research_session, research_export). Search LENSES restrict results to trusted domains (academic, clinical, legal, finance, government, journalism, devops, docs). Pluggable backends (Google PSE/Brave/Serper/SearXNG/SearchAPI/Exa), DuckDuckGo fallback. Use for reputation-attached research needing verifiable citations — the open, auditable, private counterpart to perplexity-research (fast closed synthesis). NOT for quick uncited lookups (use perplexity/ceramic), interactive browser flows or JS-rendered pages (headless scrape tier DISABLED here → delegate to the `browser` sidecar), or grounding in our own ontology (use ontology-augment).
Production-grade live-web research bridge: search across the engine you choose,
narrow to the sources you trust (lenses), read the full source (not snippets),
and get citations you can verify — never fabricated, never a closed pre-synthesized
garden. Runs locally/private; never invokes a browser from inside this skill.
Which web-search skill? (read this first — avoids the common mix-up)
Search priority: 1. ceramic → 2. perplexity → 3. Claude WebSearch
Complex/important? → all three in parallel
Verifiable citations needed? → web-researcher (this)
Need
Skill
Default web search — keyword, rich 8k extracts, fast
Verifiable, reputation-attached research — real citations, full sources, you control which domains are searched (lenses), private/local; client work, filings, publications, medical/legal/finance
web-researcher (this)
Multi-agent deep report — fan-out + adversarial verification + cited synthesis (orchestrates searchers; can use any of the above as a backend)
deep-research
Expand a single known URL
scrape_page (this skill, full text/PDF/YouTube) ; JS-rendered → browser sidecar
Interactive browser (login/click/JS render)
browser / playwright
One-liner: ceramic = primary keyword search with dense extracts; perplexity = secondary, synthesized + authoritative; web-researcher = verifiable citations + lenses; deep-research = the harness over all.
When To Use
Research where your reputation is on the line — citations must be real and checkable.
Restrict search to trusted sources via (PubMed/arXiv/SEC/.gov, not random blogs).
lenses
Read full articles — web pages, PDF/DOCX/PPTX, YouTube transcripts, Hacker News threads.
Academic work: real papers + DOIs (academic_search), citation neighborhoods (citation_graph).
Verify a citation before relying on it (verify_citation) or audit a whole reference list (audit_bibliography); snapshot a source to Wayback (archive_source); export a bibliography (format_bibliography).
Curated resource discovery (awesome_list_search) or a structured company/brand pass (brand_research).
Session-tracked multi-step digs (sequential_search) with recovery + export.
When Not To Use
Quick casual lookup, no citing needed → perplexity-research or Claude built-in search.
Single known URL summary → scrape_page (this skill); JS-rendered → browser sidecar.
Grounding in OUR formal ontology/KG → ontology-augment.
Connection
MCP stdio server registered as web-researcher in mcp/mcp.json. Deployed binary is
web-researcher-mcp-1.43.0 (Nix-baked, resolved from $PATH); the full v1.33+
toolset — the 21 tools tabled below — is registered and live, verified against the
running server's tools/list. Registration, version-bump (flake.nix pin) and
provider-gating detail: references/deployment.md.
Tools (live set depends on configured providers — see notes)
Search & read
Tool
What it does
web_search
Search the web; optionally restrict to trusted sources via a lens
search_and_scrape
Search then read the best results, quality-scored + deduped
scrape_page
Read any URL in full — pages, PDF/DOCX/PPTX, YouTube transcripts, HN (API); mode:raw for verbatim
image_search
Images by size/type/colour/format
news_search
Recent news with date controls + source filtering
sequential_search
Multi-step research that remembers prior findings
awesome_list_search
Mine curated GitHub "awesome-*" lists for a topic's canonical resources
brand_research
Structured research pass on a company/brand
Domain search
Tool
Source
academic_search
Real papers + DOIs (OpenAlex/Semantic Scholar/arXiv/PubMed/IEEE)
citation_graph
Walk a paper's citing/cited neighborhood — needs a citation-capable provider
patent_search
US/EP/WO/JP/CN/KR patent offices + classification
legal_search
US court opinions/dockets via CourtListener
clinical_search
ClinicalTrials.gov (discovery, not medical advice)
econ_search
World Bank indicators (keyless) + FRED US macro series
Citation integrity & output
Tool
What it does
verify_citation
Does a citation exist, match a real record, is it retracted/dead-link? Evidence, not a verdict
audit_bibliography
Audit a whole CSL-JSON/RIS/BibTeX list/session — per-entry + corpus flags
verify_recommendation
Check a recommended source before relying on it
archive_source
Capture a fresh Wayback snapshot so a cited page stays verifiable (write tool)
format_bibliography
APA / MLA / BibTeX / RIS / CSL-JSON (Zotero/EndNote/Mendeley-ready)
Session / memory
Tool
What it does
get_research_session
Recover a session after context loss
research_export
Export a provenance-tracked report (md/JSON)
All 21 tools above are live in the deployed v1.43.0 binary (verified via tools/list).
Some are inert without provider keys — see references/deployment.md.
docs/TOOLS.md upstream is the authoritative, CI-verified tool list + schemas.
Search Lenses (the differentiator)
Lenses restrict results to a curated set of trusted domains for a field, instead of the
whole web. Built-in lenses ship with the binary (catalog at lenses://catalog):
academic, academic-extended, clinical, legal, finance, government,
journalism, devops, docs (+ custom JSON lenses). Reference by short name:
Live status/diagnostics the agent can read: stats://tools, stats://sessions,
stats://rate-limits, stats://providers, lenses://catalog,
diagnostics://errors/recent, diagnostics://health, and a large-payload artifact
store research://artifact/{id}. Ready-made research prompt templates appear as / commands.
Browser Delegation (Critical)
web-researcher-mcp's tier-4 scrape (go-rod + stealth) would auto-download its own
Chromium per pod. In agentbox this is disabled — a second Chromium duplicates the
browsercontainer sidecar (Chrome Beta 149+, NVIDIA Vulkan, VNC :5903, MCP SSE :8931):
SCRAPER_DISABLE_BROWSER=true
CHROME_PATH= # empty -> tier 4 hard-fails fast
When tiers 1–3 (markdown negotiation, stealth HTTP, HTML parser) can't extract a
JS-rendered page (NEEDS_BROWSER), switch to the browser skill — do not retry inside this MCP:
1. scrape_page(url) -> tiers 1-3 ok => done ; all fail => NEEDS_BROWSER
2. browser_navigate({url}) on browser-gpu
3. browser_snapshot() / browser_evaluate(...)
One Chrome surface, one GPU allocation, one audit trail.
Required Configuration
Set at least one search provider (read from session env):
Note: gemini-url-context / web-summary are referenced by some older skill docs but
the backing gemini CLI is not on PATH here — those routes are non-functional
pending a gemini CLI install. Use scrape_page (this skill) or the browser sidecar
for single-URL expansion instead.
