Every search tool returns sources; almost none of them tell you where the sources disagree. That is the actual research problem — three pages give three numbers and someone has to notice. Running several engines concurrently matters for the same reason, since one engine's ranking is one opinion about what is relevant. Being hosted with no key is what makes it worth trying at all: connect, ask one question, and decide.
A FastMCP server exposing nine tools that search, fetch, score and compare information across the open web, social platforms, news, academic literature and code hosts. It is hosted, so connecting takes a URL and nothing else. Every tool carries behaviour annotations, so a client can tell read-only calls from the rest and skip confirmation prompts accordingly.
- web_search — several engines run concurrently, including DuckDuckGo, Bing, Yahoo, Mojeek and Wikipedia, with TLS fingerprinting that gets past bot walls a plain fetch would hit
- github_search for code and repositories, social_search across social platforms, and news_aggregation for current coverage
- map_website — walk a site's structure rather than fetching one page at a time
- content_operations — one tool covering six operations: retrieve, stream, analyze, extract, score and find_conflicts, that last one comparing sources against each other rather than summarising each
- document_analysis and entity extraction, for what comes back
- research_topic — a workflow rather than a query, chaining search and retrieval into a topic write-up
- scientific_research — academic papers and datasets across five paper providers, including OpenAlex, CrossRef, arXiv, PubMed and Europe PMC, plus four dataset hubs
- No model runs inside the server: it retrieves and structures, your assistant does the reasoning
Nothing to connect to the hosted server — point a client at https://RivalSearchMCP.fastmcp.app/mcp; no account, no key, no rate-limit paperwork. Running it yourself needs Python and uv, and the FastMCP CLI can install it into Claude Desktop or run it over stdio or HTTP. It is also published as a plugin for Claude Code and OpenAI Codex, and as a standalone agent skill that uses the same nine tools through a CLI without any MCP configuration.
One command — claude mcp add RivalSearchMCP --url https://RivalSearchMCP.fastmcp.app/mcp