Firecrawl Blog

Firecrawl, Playwright, Composio, Asana, Context7, Slack, Supabase, and Filesystem: the MCP servers I actually keep configured in Gemini CLI, with copy-pasteable settings.json snippets.

Recently, Kimi K3 was released by Moonshot AI. In this guide, we'll go over Kimi K3 specs, providers, and how to use it with OpenCode.

Agentic OCR wraps traditional OCR in an AI agent that reviews, corrects, and re-runs extraction until the output is clean. Here's how it works, plus a working Python agent built on Firecrawl.

We've put together a list of 20 AI automations for business workflows. Most are small automations that offer real productivity gains. Read on to see which one is right for your team.

Twelve data extraction tools I actually reach for in 2026, across web, documents, databases, and no-code, with honest takes on each.

Firecrawl, Slack, Asana, Canva, ElevenLabs, Runway, and Hex: the ChatGPT plugins and MCP connectors that actually change how I use ChatGPT in 2026.

How amotivv relies on Firecrawl's /scrape, /search, and /map to give AI agents unconditional retrieval of live web content for regulated business customers.

Two Firecrawl Rust libraries that turn every common document format into clean markdown. pdf-inspector handles PDFs. AnyDoc handles the other 14 formats. Both now power /parse and /scrape.

Data is the oil of the 21st century. Everything runs on data. Not all web data is created equal. For production systems, we need fresh, high-quality web data. Read to learn which is best for your team.

Firecrawl's official ChatGPT plugin gives ChatGPT direct access to live web data. Run real web searches with full-content results, scrape any page or document, crawl entire sites, interact with dynamic pages, and monitor the web for changes without leaving the conversation.

Claude Code expects Claude models. CLI Proxy is a local proxy server that translates API formats both ways, so you can point Claude Code at GPT, Gemini, or Grok and keep skills, subagents, and reasoning efforts working.

Firecrawl, ChatGPT, PDF.ai, ChatPDF, Smallpdf, and TLDR This ranked on how they actually summarize PDFs, plus which ones expose an API, MCP, or CLI for AI-first workflows.

Twelve Antigravity skills tested in a clean workspace: Firecrawl for live web data, Build with Google bundles, and cross-agent picks.

Compare LangChain, Microsoft Agent Framework, CrewAI, Agno, Haystack, LlamaIndex, and the OpenAI Agent SDK on multi-agent support, languages, license, and hosting.

Every time Claude Code scrapes or searches the web mid-session, that page's content sticks around in context for every message after it. Four Firecrawl techniques, tested live against Claude Code's default tools, cut what actually lands in context.

GLM-5.2 vs Kimi K2.7 Code across scraping, calculator build, refactor, and Fibonacci. GLM wins on runtime and refactors; Kimi wins on clean algorithm code.

Contentful uses Firecrawl to turn any prospect's website into structured content for DemAI, an AI demo app that builds personalized Contentful demos.

Firecrawl is now an official Replit Connector. Add state-of-the-art web search, scraping, and browser interaction to your apps in a single click.

Firecrawl /search now uses a new custom relevance model to return the excerpts that best answer your query. Agents using /search score 94.7% on SimpleQA while using 10x fewer tokens than processing full pages.

WebMCP lets a site expose developer-defined tools to an AI agent, but only inside a live browser tab. Firecrawl's interact endpoint runs a real browser in the cloud, so headless agents like Claude Code can call those tools too.

Firecrawl's monitor endpoint scrapes a page on a schedule, diffs it against the last snapshot, and uses an AI judge to only notify you when a change is actually meaningful.

Most agents grab page one of a paginated site and call it done. Here's how to walk every page with the Firecrawl SDK and know exactly when to stop.

How Minimal uses Firecrawl to process nearly six million storefront pages a month, turning live product and policy content into data for AI support agents.

The best AI for research in 2026, from real-time answer engines to peer-reviewed literature search to the retrieval infrastructure powering research agents. Verified capabilities, honest tradeoffs, and where each of these top AI tools for research fits in a modern stack.

Docling, PyMuPDF, pdf2json, pdf-parse, and Firecrawl compared for converting PDF files to structured JSON.

The best OpenCode skills for extending Anomaly's model-agnostic coding agent, from live web context with Firecrawl to multi-agent orchestration with Superpowers and codebase understanding with knowledge graphs.

Looking for the best Nimbleway alternatives in 2026? Compare Firecrawl, Bright Data, Oxylabs, Apify, and Zyte on pricing, output format, and AI features.

Firecrawl /monitor now watches the entire web, pinging you or your agent the moment something relevant comes online.

AI agent architecture is an LLM wrapped in a harness of tools, memory, a knowledge base, and orchestration. These components turn a model into an agent.

A builder's shortlist of the five AI assistants worth using in 2026: Vellum, Manus, Claude Cowork, OpenClaw, and Hermes Agent, scored on real work rather than demos.

The Hermes Agent plugins worth installing in 2026: Firecrawl for live web context, Composio for app connectivity, Honcho and Hindsight for memory, Langfuse for observability, and the Kanban dashboard for multi-agent work.

Build a competitor monitoring system with Firecrawl Monitor: schedule scrapes, diff changelog and pricing pages, and get AI-filtered alerts in one API.

A web crawler is a program that follows links to discover and collect web pages. Here is how the crawl loop works, the policies that govern it, and how to run one for an AI agent.

Build a Slack AI agent with eve, Vercel's filesystem-first framework, and wire in Firecrawl web search for cited answers in any thread.

Firecrawl skills give Claude Code and other agents new abilities, not just a new personality: search, scrape, market research, SEO audits, and design system cloning from one sentence.

The Grok plugins worth installing in 2026: Firecrawl for live web context, Composio for app integrations, Vercel for deploys, Sentry for production errors, and Matt Pocock's skill pack for engineering judgment.

Puppeteer is a Chrome-focused Node library that drives the browser over the DevTools Protocol. Selenium is the cross-browser, multi-language WebDriver standard. Here is how they differ, the same task built in both, and how to pick.

Flue (Astro/Cloudflare) and Eve (Vercel) are new agent frameworks that converged on the same shape. Plus how to add Firecrawl web search to each.

Which ScrapingBee alternative fits depends on the use case: AI-ready output, proxy scale, marketplace coverage, or no-code scraping. Five tools compared.

Cursor Automations deploy AI agents to the cloud with one toggle. Here's how to wire one up with Firecrawl for live web access.

Firecrawl Research Index gives agents the entire AI/ML literature and the code behind it, with SOTA recall on arXivQA — 18% above the next best provider at comparable cost.

A web search API gives your AI agent a programmatic way to query the live web and get back structured, LLM-ready results. Here's how it works, what to look for, and how to call one.

Firecrawl is now keyless. Get 1,000 free credits a month to search, scrape, and interact with the web. No account, no setup, no key rotation.

The Codex plugins worth installing in 2026 — from Firecrawl for live web context to Composio for app integrations, Context7 for fresh docs, Superpowers for agent workflows, and RTK for token compression.

Loop engineering is the shift from prompting coding agents one turn at a time to designing the system that prompts them. Here's what a loop actually is, the parts it needs, and how to build one that halts.

Why AI agents prefer the CLI: the token math, Terminal-Bench data, and the Unix tools (git, gh, ripgrep, curl) agents reach for.

A sourced comparison of the 8 best AI coding agents in 2026, ranked on harness depth, remote agents, token cost, and benchmark accuracy.

How to run multiple Codex agents in parallel using subagents, custom TOML agents, git worktrees, and Symphony — with a runnable three-agent review workflow.

Explore the best Parallel AI alternatives for AI web search and data extraction. Compare Firecrawl, Exa, Tavily, and Linkup across accuracy, pricing, and capabilities.

How to cut Claude Code token consumption by 77-91%: verified benchmarks on CLAUDE.md trimming, path-scoped rules, model routing, MCP overhead control, and clean web data via Firecrawl.

Claude Code has the deepest harness and fastest terminal. Codex covers more surfaces at a lower price. Compared on pricing, sandboxing, benchmarks, and community verdict.

OpenClaw vs Hermes Agent compared on architecture, memory, file handling, security, and deployment, tested side by side in isolated Docker containers on the same model.

Anthropic's web_search billed 232,015 input tokens in a single turn. Four tested alternatives: Firecrawl, Brave, Exa, and Parallel, with install commands and live test results.

The best low-code AI workflow automation tools in 2026: n8n, Cursor Automations, Claude Routines, Zapier, Make, Gumloop, and Vellum. Visual builders and AI-native platforms for teams who want to build AI pipelines without heavy coding.

Compare six search APIs for AI agents on latency, pricing, and built-in content extraction: Firecrawl, Brave, Exa, Tavily, Perplexity Sonar, and Serper.

AI SDR tutorial on how to build a Python agent that researches any company live with Firecrawl web search, pulls structured intel from its site, writes a specific outreach message, then scales across a list.

Claude Code is managed and model-locked; OpenCode is open-source and model-agnostic. Compare cost, terminal UX, harness depth, and the current state of the subscription block.

Firecrawl Monitoring watches pages for you and notifies your agent the moment something changes. Use up to 90% fewer LLM tokens by only ingesting what actually changes on a page.

The best GitHub repos for AI developers in 2026. Twelve repositories covering supply chain scanning, web context APIs, LLM training fundamentals, agent skill packs, CSS design skills, and more.

Firecrawl is now a native integration on the Vercel Marketplace. One-click install, API key auto-injected into your project environment, billed through Vercel.

LLM agents hallucinate from stale data and knowledge gaps. Why hallucinations persist despite smarter models, and how real-time web search reduces them.

A finance research agent that scrapes SEC filings, IR pages, and financial news live with Firecrawl, then passes clean data to an LLM for analysis. Full TypeScript code included.

The best Codex skills for extending what OpenAI's coding agent can do, from live web context with Firecrawl to PR automation, MCP server scaffolding, and more.

Firecrawl covers Search, Scrape, Crawl, Map, Interact, and Parse. Here's which endpoint to reach for and how to compose them in an agent workflow.

/interact keeps a browser session alive between API calls. Drive it in plain English or code to handle logins, multi-step flows, and lazy-loaded data. Covers prompt mode, code mode, and named profiles.

A practical comparison of the best document parsing APIs for developers building AI pipelines, RAG systems, and document automation workflows in 2026.

The best investment research APIs in 2026 for alternative data, fundamental analysis, real-time market feeds, and news sentiment. Tested and compared.

Agentic search queries the live web at reasoning time, making it the right retrieval method for real-time signals and open-web content that RAG can't cover.

Two new formats for Firecrawl /scrape — question returns grounded answers from any web page, highlights returns verbatim excerpts. Both use up to 100x fewer tokens than a full scrape, with a fully managed LLM stack and prompt-injection hardening built in.

Codex's built-in web search queries a stale pre-scraped cache by default. Learn how to switch to live mode and wire in Firecrawl MCP for real-time results, full-page content, and crawling.

LLM grounding injects factual, up-to-date context at query time so the model reasons over real data. Covers the search-scrape-inject pipeline with working Python code.

Hermes ships Firecrawl as its default web backend. Here's what web_search, web_extract, and the browser tools actually do, how they differ, and the patterns that work for long-running agent tasks.

Lockdown Mode is Firecrawl's new cache-only scrape mode - no outbound request, zero data retention, one flag across every SDK, the CLI, and MCP.

Upload a PDF, Word doc, or spreadsheet and get back clean markdown, a summary, or structured JSON in one call.

Training data vs. retrieved data vs. live web data, learn when each layer matters for AI agents and how to combine them for accurate, current answers.

Agent skills are reusable SKILL.md files that give AI coding agents specialist knowledge on demand. Learn how they work, why they went from 0 to 40,000 in 20 days, and how to build your own.

A hands-on comparison of the best PDF parsers for AI and RAG pipelines in 2026, covering speed, output quality, table handling, and LLM-readiness for each tool.

Comparing the top Bright Data alternatives for web scraping and SERP APIs in 2026: real pricing, feature breakdowns, and community-sourced context.

A developer's guide to the best news API available in 2026: real-time news search, article extraction, and AI-ready data pipelines for monitoring, research, and agent workflows.

Hermes is a local-first agent framework from Nous Research that learns from past tasks, carries memory across sessions, and ships with Firecrawl as its default web backend. This guide covers what makes it different, how to install it, and how to extend it with web search, scraping, and crawling.

A web index is a pre-built snapshot of the web that powers both search engines and AI agents. Covers the four-stage pipeline, hybrid retrieval, and why index quality limits what your agent can answer.

Give Claude, Cursor, Codex, or any MCP client real-time web access in minutes. We compare the top web search MCP servers (Firecrawl, Tavily, Exa, and WebSearch-MCP) with exact install commands and honest takes on where each one falls short.