Web Research MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| JINA_API_KEY | No | Jina API key. Provides higher fetch rate for fetch_url (free tier: 1M tokens/month). | |
| BRAVE_API_KEY | No | Brave Search API key. Enables general web search via search_web (free tier: 2,000 queries/month). | |
| TAVILY_API_KEY | No | Tavily API key. Enables general web search and research-optimized snippets via search_web (free tier: 1,000 queries/month). | |
| WEB_RESEARCH_MAX_RETRIES | No | Maximum number of retries for transient network errors, HTTP 5xx responses, and HTTP 429 responses. | |
| WEB_RESEARCH_TIMEOUT_SECONDS | No | Default timeout in seconds for provider HTTP requests. | |
| WEB_RESEARCH_MAX_BACKOFF_SECONDS | No | Maximum backoff seconds between retries. | |
| WEB_RESEARCH_TIMEOUT_JINA_SECONDS | No | Provider-specific timeout override in seconds for Jina requests. | |
| WEB_RESEARCH_RETRY_BACKOFF_SECONDS | No | Base backoff seconds between retries. | |
| WEB_RESEARCH_TIMEOUT_WIKIPEDIA_SECONDS | No | Provider-specific timeout override in seconds for Wikipedia requests. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| search_webA | Multi-source web search. Aggregates results from Brave Search and/or Tavily (whichever has a key configured) with dedup and cross-source scoring. Returns title, URL, snippet, and source for each result. Use as the default 'search the web' tool. |
| fetch_urlA | Fetch any URL and return clean markdown content. Uses Jina Reader which handles JS rendering and bot detection on your behalf, returning readable text. Ideal for reading articles, papers, docs, or blog posts. |
| search_wikipediaA | Search Wikipedia. Returns titles, URLs, and snippets. No API key required. |
| search_academicA | Search arXiv for academic preprints across all fields (CS, physics, math, bio, etc.). Returns title, authors, abstract snippet, and PDF URL. No API key required. |
| search_newsA | Search Hacker News for tech news, discussions, and trending links. Returns title, URL, points, comments, and dates. No API key required. |
| search_stackexchangeA | Search Stack Exchange sites (default: Stack Overflow) for high-quality technical Q&A. Set the 'site' parameter to search other SE communities (e.g. 'serverfault', 'superuser', 'askubuntu', 'math', 'tex'). No API key required. |
| search_scholar_metaA | Search Crossref for scholarly metadata across publishers (Elsevier, Springer, Wiley, IEEE, ACM, etc.). Returns DOI, citation count, publication date, and abstract. Covers peer-reviewed papers that arXiv may not. No API key required. |
| plan_researchA | Build a structured research plan for a complex question without executing it. Returns the planned sub-questions, the sources recommended for each, and estimated cost (searches + fetches). Use this when you want to review or edit the plan before committing to the full deep-research pipeline. Returns JSON; pass it back to the |
| extract_evidenceA | Fetch a URL and extract the passages most relevant to a specific question. Returns each passage with a short context window before/after, a relevance score, and a character offset into the original page so citations are verifiable. Use this when you want to drill into a specific source for evidence on a narrow claim. |
| researchA | Run a full deep-research pipeline on a complex question. Decomposes the question into sub-questions, fans out across multiple sources (Wikipedia, arXiv, Hacker News, Stack Exchange, Crossref, plus Brave/Tavily if keys are configured), fetches the top URLs, extracts the most relevant passages, and returns a structured ResearchReport containing: the plan, a numbered citation manifest, per-sub-question evidence with quotes + character offsets, and a Markdown synthesis template for you to fill in. You (the model) should write the narrative synthesis citing the [n] markers; the server does the gathering, not the writing. Use |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 10 tools
Most tools have clear, distinct purposes: general web, Wikipedia, arXiv, Hacker News, StackExchange, Crossref metadata, and evidence extraction are clearly separate. Minor overlap exists between search_academic (arXiv preprints) and search_scholar_meta (Crossref metadata), since both find scholarly papers, but their descriptions make the distinction clear enough. The plan_research and research tools are also clearly separated as planning vs. execution.
Tools mostly follow verb_noun naming (search_web, fetch_url, search_wikipedia, plan_research, extract_evidence), but some are slightly off: search_scholar_meta uses a noun descriptor for a source rather than a clean object; research is a standalone verb, and fetch_url/extract_evidence are both verb_noun but diverge in verb choice. No mixed casing or chaotic naming, so overall predictable and readable.
10 tools is within the ideal 3-15 range and appropriate for a research server: search methods for multiple sources, plus a planner, an executor, and a fetcher/extractor. Each tool serves a distinct role without redundancy, and the scale is not too heavy to handle.
The server covers the core research pipeline fairly well: plan, search, extract, synthesize. Minor gaps: no full-text search on Crossref tokens or thousands; abstracts only for arXiv; no search by author/full-text Google Scholar type search; PDF retrieval across fetch_url may not work; no citation graph or retrieval. Useful but not complete.