Research & Writing
Five MCPs for the find → ingest → write loop used by researchers, analysts, and content marketers in 2026: Tavily for live web search (1,917 stars), Firecrawl for clean URL-to-Markdown extraction (6,229 stars), Apify for structured scraping at scale (1,190 stars), Microsoft's Markitdown for converting local PDFs and Office files (121,110 stars), and Notion as the destination (4,302 stars).
Use case: For researchers, analysts, and content marketers whose week is "find sources → read → publish." Tavily handles live questions ("what's the current data on X"), Firecrawl converts a known URL into clean prose, Apify scrapes structured datasets that Firecrawl can't, Markitdown ingests local PDFs and decks, and Notion is where the writeup ends up.
“Markitdown converts PDFs, Word docs, Excel sheets, PowerPoint files, images, audio, HTML, and more into Markdown. Useful for getting messy file content into an LLM context window. — microsoft/markitdown README”
What's in the stack
MCP servers and CLI tools that make up this stack, and how each one runs.
| Tool | Primary use | Kind | Stars | GitHub |
|---|---|---|---|---|
| Tavily connects an agent to real-time web search, page extraction, site mapping, and systematic crawling through the Tavily API.. | MCP | 1,917 | repo | |
| Firecrawl searches the web, scrapes URLs into clean structured data, crawls multi-page sites, and can run autonomous deep-research sessions with a browser agent.. | MCP | 6,229 | repo | |
| Apify MCP Server exposes thousands of pre-built web scrapers and automation Actors, letting an agent extract structured data from websites, social platforms, e-commerce stores, and search engines.. | MCP | 1,190 | repo | |
| Markitdown converts PDFs, Word docs, Excel sheets, PowerPoint files, images, audio, HTML, and more into Markdown. | MCP | 121,110 | repo | |
| Notion MCP Server connects an agent to a Notion workspace: search and read pages, create and update content, manage databases, and move pages between locations.. | MCP | 4,302 | repo |
Install the MCP servers
Pick the client you use. Each tab installs all 5 servers.
claude mcp add io-github-tavily-ai-tavily-mcp -- npx -y tavily-mcp
claude mcp add firecrawl-firecrawl-mcp-server -- npx -y firecrawl-mcp
claude mcp add --transport http com-apify-apify-mcp-server https://mcp.apify.com/
claude mcp add microsoft-markitdown -- uvx markitdown-mcp
claude mcp add --transport http makenotion-notion-mcp-server https://mcp.notion.com/sseAdd to your CLAUDE.md / AGENTS.md
Paste this into your project's CLAUDE.md or AGENTS.md so the agent knows when and how to use each tool in the stack.
## Tools
- **Tavily MCP** — Use for live questions where freshness matters: current statistics, recent announcements, "what's the latest on X." Cite the URL Tavily returns; do not summarize without attribution.
- **Firecrawl MCP** — When the user provides a URL or you've found one via Tavily, extract it with Firecrawl into clean Markdown before reading. Far cheaper than asking the model to read raw HTML.
- **Apify MCP** — For structured scraping (LinkedIn profiles, Instagram posts, product catalogs, anything Firecrawl can't extract cleanly). Pick the right Apify Actor for the site, then run it with the smallest possible result limit first.
- **Markitdown MCP** — For local files (PDFs, Word, Excel, decks, images, audio). Convert before reading; never paste raw binary or extract text fragments by hand.
- **Notion MCP** — Drafts and findings live here. Always create as a draft page and ask the user to review before publishing or moving into a shared database.Why each tool is in this stack
A short rationale per tool, plus its individual install.
Tavily's official MCP (1,917 stars). Real-time web search, page extraction, site mapping, and systematic crawling. Use Tavily for "what's current" questions where freshness matters more than depth.
claude mcp add io-github-tavily-ai-tavily-mcp -- npx -y tavily-mcpFirecrawl's official MCP (6,229 stars). Scrapes individual URLs and multi-page sites into clean structured Markdown. Ideal once you know the source you want — paste a URL, get readable content with no boilerplate.
claude mcp add firecrawl-firecrawl-mcp-server -- npx -y firecrawl-mcpApify's official MCP (1,190 stars). Exposes thousands of pre-built scrapers ("Actors") for sites Firecrawl can't extract cleanly — e.g., LinkedIn, Instagram, structured marketplaces. Used in production by Josh Lachkovic at Ballpoint for ad creative scraping and creator discovery.
claude mcp add --transport http com-apify-apify-mcp-server https://mcp.apify.com/by microsoft · ★ 121,110 · MCP
Microsoft's Markitdown (121,110 stars). Converts PDFs, Word, Excel, PowerPoint, images, audio, and HTML into Markdown locally — no upload, no third-party service. The bridge between a folder of source files and a chat that can actually read them.
claude mcp add microsoft-markitdown -- uvx markitdown-mcpby makenotion · ★ 4,302 · MCP
Notion's official MCP (4,302 stars). Search and read pages, create and update content, manage database items. The destination for synthesized findings — agents draft, you review.
claude mcp add --transport http makenotion-notion-mcp-server https://mcp.notion.com/sse
Last reviewed · Curated by stax.sh