SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsOther AI toolsRivalSearchMCP

RivalSearchMCP

RivalSearchMCP is a deterministic research MCP server based on FastMCP 3.x, enabling AI agents to use multi-source search tools including web, social, academic, news, code, and documentation through a single connection. It provides 9 specialized tools, covering 5 search engines, 9 social platforms, 5 news sources, 5 academic databases and 4 dataset hubs, as well as GitHub repository search and document analysis. All tools require no API keys, the server does not run an LLM internally, and it returns deterministic, auditable structured output, automatically accompanied by quality scores and conflict detection signals, making it convenient for agent chaining. The project is MIT-licensed, open source and free, and can connect to the official hosted server or be self-deployed.

Open website

About this tool

RivalSearchMCP is a deterministic research MCP server based on FastMCP 3.x, designed to provide AI agents with unified multi-source information retrieval capabilities. Its core features are no API keys, no LLM running inside the server, and all tools returning deterministic, auditable output, with synthesis performed by the caller's model.

Purpose

RivalSearchMCP is used to let AI assistants or agents complete cross-source research work in a single connection, including web search, social discussion retrieval, news aggregation, academic paper and dataset search, GitHub repository lookup, website traversal, and document analysis. It is suitable for scenarios that need to collect, score, and compare information from multiple independent sources, while avoiding configuring a separate MCP server for each source.

Main Capabilities

  • Search and Discovery (5 tools)

    • web_search: concurrent multi-engine search across DuckDuckGo, Bing, Yahoo, Mojeek, and Wikipedia, with results deduplicated, TLS fingerprint-secure scraping via Scrapling, and the ability to bypass Cloudflare/Akamai front-line protection. Failure of a single engine does not affect other engines.
    • social_search: covers 9 platforms in total, including Reddit, Hacker News, Stack Overflow, Dev.to, Medium, Product Hunt, Bluesky, Lobste.rs, and Lemmy, with no authentication required.
    • news_aggregation: aggregates 5 news sources in total, including Google News, Bing News, The Guardian, GDELT, and DuckDuckGo News, and supports time_range filtering (day/week/month/anytime).
    • github_search: repository search with built-in rate limiting (60 requests/hour unauthenticated), and optionally includes README.
    • map_website: traverses websites in research, docs, or map mode, returning per-page quality scores and aggregated confidence signals.

  • Content Analysis (3 tools)

    • content_operations: a single tool providing six operations—retrieve, stream, analyze, extract, score, find_conflicts. Among them, score rates URLs by tier, freshness, corroboration, and citations (0-100), returning markdown and structured JSON; find_conflicts compares 2-10 sources, detecting divergences in values, dates, and polarity and providing confidence weights.
    • research_topic: two modes—topic mode searches, crawls, and ranks key findings by relevance; entity mode fans out in parallel for named entities to web, news, GitHub, social, and academic sources, returning a unified report and confidence.
    • document_analysis: extracts text from PDF, Word, plain text, and images; images use EasyOCR (lazy-loaded, no setup required); file limit is 50 MB.

  • Research workflows (1 tool)

    • scientific_research: academic paper and dataset search. The 5 paper providers are OpenAlex, CrossRef, arXiv, PubMed, and Europe PMC; the 4 dataset hubs are Kaggle, HuggingFace, Dataverse, and Zenodo.

Each tool includes ToolAnnotations (readOnlyHint, openWorldHint, destructiveHint, idempotentHint) so MCP clients can skip confirmation prompts when safe; each tool has a timeout cap to prevent hanging sources from blocking the client. Multi-result tools automatically include per-item quality scores (0-100) and aggregate confidence signals.

Usage

RivalSearchMCP runs as a remote MCP server hosted on FastMCP, with no need to clone a repository or run a local server.

  • Connect to the hosted server

    • Claude Desktop: Settings → Add remote server, enter the URL https://RivalSearchMCP.fastmcp.app/mcp
    • VS Code: add the configuration to the .vscode/mcp.json file
    • Claude Code: use the built-in MCP management command claude mcp add RivalSearchMCP --url https://RivalSearchMCP.fastmcp.app/mcp

  • Local installation

    • Offers one-click installation via the FastMCP CLI, installation-free quick run, development with the Inspector, and manual UV setup, among other methods.

  • Plugins and skills

    • Can be used as an installable plugin for Claude Code and OpenAI Codex. The GitHub repository also serves as a plugin marketplace; after adding the marketplace, install rival-search-mcp (registers the hosted MCP server and exposes 9 tools) and/or rival-search-mcp-skills (installs standalone agent skills, reference documentation, and CLI helpers).
    • Provides a Claude Code Agent Skill: a self-contained CLI that enables AI agents to use all 9 tools without MCP configuration. After copying the skill to the Claude Code skills directory, Claude will automatically discover and use it when web research, competitive analysis, or content discovery is needed. The CLI can also be run directly via uv, with inline dependencies, fully self-contained.

After installation, Claude Code exposes the MCP tools as mcp__RivalSearchMCP__*, and example prompts can be used such as: "Use RivalSearchMCP to research FastAPI vs Django. Run research_topic on both, aggregate recent news, check Reddit and Hacker News discussions, search GitHub for activity, look for academic papers, score the top sources, and flag any conflicts between them."

Use cases

  • Research tasks that require collecting and comparing information across multiple sources such as the web, social media, news, academic sources, and code.
  • Tool outputs that need to be structured and chainable in agent workflows.
  • Scenarios that require detecting conflicts in values, dates, or positions between different sources.
  • Scenarios that require quickly scoring source quality and assessing confidence.
  • Scenarios that require extracting text from documents (including image OCR) and analyzing it.
  • Situations where you want to work without API keys and complex configurationusers who use MCP clients in such cases.

Limitations

  • No LLM runs inside the server, and no autonomous agent loop is provided; if an autonomous loop is needed, it must be implemented on the client side.
  • The hosted server enforces fair-use rate limits; if different limits are needed, you can self-host.
  • GitHub search is rate-limited to 60 requests/hour when unauthenticated.
  • The document analysis file limit is 50 MB.
  • The core tools are free and open source (MIT license), but the documentation does not provide paid tier or pricing information.
  • Works with any MCP-compatible client, including Claude Desktop, Cursor, VS Code, and Claude Code; actual platform compatibility is subject to testing.

Related tools

Browse category