MCP Omnisearch is a Model Context Protocol server that puts several search and extraction providers behind four tools, so an agent can query Tavily, Brave, Kagi, Exa, GitHub, Linkup, and Firecrawl without a separate server for each. The agent picks a provider per call, and providers without a configured key are skipped while the rest keep working.
The four tools split the work by task. Web search queries Tavily, Brave, Kagi, Exa, or Kagi Enrichment with controls such as depth, topic, time range, and safe search where the provider supports them. AI search returns sourced answers from Kagi FastGPT, Exa Answer, Linkup, or Tavily Research, the last returning a task id that is passed back to fetch the report. GitHub search covers code, repositories, and users. Web extraction extracts, crawls, scrapes, summarizes, or finds similar content through Tavily, Kagi, Firecrawl, or Exa, and Firecrawl can point at a self-hosted instance.
Large results can be returned inline or written to a file. The server works with Claude Code, VS Code, and other MCP clients, and provider keys can be referenced through the client's secret expansion rather than stored in plain text.
Features
- Web search: one tool across Tavily, Brave, Kagi, Exa, and Kagi Enrichment with depth, topic, time-range, and safe-search controls
- AI answers: sourced answers from Kagi FastGPT, Exa Answer, Linkup, or Tavily Research
- GitHub search: search code, repositories, or users
- Content extraction: extract, crawl, scrape, summarize, or find similar pages through Tavily, Kagi, Firecrawl, or Exa
- Graceful provider fallback: providers without keys are skipped and the others stay available
- Large-result handling: results returned inline or written to a file
- Self-hosted Firecrawl: extraction can target a custom Firecrawl instance
