TinyFish reads URLs.
Firecrawl runs the whole pipeline.

Both Firecrawl and TinyFish give AI agents live web data, but the reach and the openness of the two products are different. Firecrawl covers the full pipeline: whole-site /crawl, /map for URL discovery, /batch/scrape for concurrency, /monitor for scheduled change detection, structured extraction on /scrape, /parse for uploaded files, and two purpose-built indexes on top (43M+ paper abstracts, 70M+ docs and merged pull requests). The core platform is open source under AGPL-3.0 with 180K+ GitHub stars, and Firecrawl ships nine official SDKs plus a first-party MCP server used by Claude, Cursor, Windsurf, and Zed. TinyFish exposes Search, Fetch, Agent, Research, and Browser primitives for AI agents, publishes an open source cookbook, and has Python and Node SDKs; its Fetch endpoint caps at 10 URLs per request and does not include a crawler, a map, batch scale, or a first-party /monitor endpoint (TinyFish teaches customers to build monitoring themselves on top of Agent runs).

Yes. Firecrawl returns clean markdown by default, and /scrape accepts a JSON Schema or a natural-language prompt for structured extraction in the same call. TinyFish Fetch also returns clean markdown, but structured extraction is only available on Agent runs at $0.016 per step; Fetch itself has no JSON Schema output. For RAG teams doing extraction at scale, keeping schema-shaped extraction on the free /scrape tier tends to matter more than raw text cleanliness.

Firecrawl runs keyless with 1,000 free credits a month and no signup required. Paid plans (annual billing) start at $16 per month for 5,000 credits (Hobby), $83 per month for 100,000 credits (Standard), and $333 per month for 500,000 credits (Growth). One credit unit spans /scrape, /crawl, /search, /browse, and /parse. TinyFish uses a pay-as-you-go wallet model: Search and Fetch are free within rate limits, Agent runs bill at $0.016 per step, and Browser sessions bill at $0.002 per minute, drawn from a Wallet that requires a $10 minimum deposit.

Yes. The Firecrawl platform is fully open source under AGPL-3.0 with 180K+ GitHub stars, and it can be self-hosted for full control over data, compliance, and infrastructure. That matters for regulated industries, air-gapped environments, and teams that want to audit or fork the platform. TinyFish publishes an open source cookbook of examples at github.com/tinyfish-io/tinyfish-cookbook, but the platform itself is a proprietary managed service with no public source or self-host option.

Most developers are productive in minutes. Firecrawl runs keyless with 1,000 free credits a month, so an agent can hit /scrape, /search, or /parse with no signup and no API key at all; /crawl, /map, and /agent need an account. Nine official SDKs cover Python, Node.js, Go, Rust, Java, .NET, Ruby, PHP, and Elixir. TinyFish also gets started quickly with Python and Node SDKs and free Search and Fetch tiers, but paid products require a Wallet with a $10 minimum deposit. Teams that want broader language coverage or that plan to build on the LangChain and LlamaIndex-native path will find fewer bridges to write with Firecrawl.

The two products optimize for different things. TinyFish publishes strong latency numbers, including sub-250ms cold browser starts and a public web-agent benchmark for its Agent product. Firecrawl optimizes for coverage: blocked pages get retried automatically so hard-to-scrape sites still come back with content, and an internal benchmark on 1,000 diverse public URLs reached 96% coverage with a 0.638 F1 accuracy score. If p50 latency matters most, TinyFish is fast. If a corpus needs to be complete, Firecrawl retries where it counts.

It depends on the board. On OpenBenchmarks' search-only board Firecrawl ranks first of 14 at a median of 7,456 LLM tokens per task with 70.3% task completion, with TinyFish a close second at 7,469 tokens and 59.3% completion. On the search-and-fetch board, where the agent opens the pages it finds, TinyFish leads at 12,844 tokens with 79.0% completion, and Firecrawl places third at 17,379 tokens with 76.0% completion. Fewer tokens per task means a lower model bill for every agent loop.

Yes. Firecrawl's /search endpoint takes a scrapeOptions parameter so full-page markdown for every hit comes back in one call. TinyFish splits this into two products: Search returns snippets and URLs, and Fetch turns each URL into content. Both approaches work; the single-call pattern on Firecrawl is easier to wire into an agent tool and cuts latency and step count for research-heavy workflows.

Yes. Firecrawl runs keyless out of the gate: an agent can call /scrape, /search, or /parse with no signup and no API key, and add a key for /crawl, /map, /agent, and higher rate limits. The first-party MCP server plugs into Claude, Cursor, Windsurf, Zed, and other MCP-aware editors, so an MCP-aware agent picks up scrape, search, crawl, browse, parse, and the specialty indexes as tools. TinyFish is also on the MCP registry and ships integrations with ChatGPT, Claude, Cursor, Vercel, and n8n, but paid products require Wallet setup first, which adds a step to automated onboarding.

Firecrawl is a natural fit for RAG. Clean markdown is the default output, /crawl walks entire sites for corpus building, /batch/scrape processes many URLs concurrently, /monitor keeps the corpus fresh with scheduled diffs, /parse turns uploaded PDFs, DOCX, XLSX, and HTML files through the same engine, and first-party integrations exist for LangChain, LlamaIndex, Dify, and other retrieval frameworks. TinyFish focuses on live agent workflows on individual sites and does not ship a full-site crawler or a file parser, so RAG teams building over documentation, help centers, product catalogs, or uploaded documents will find fewer moving parts on Firecrawl.

Yes. Firecrawl maintains two purpose-built indexes on top of the live web API. The Research Index covers 43M+ paper abstracts across AI/ML and life sciences with state-of-the-art recall on arXivQA, and the whole index is free to use. The Developer Search Index covers 70M+ docs, READMEs, issues, and merged pull requests for coding agents, exposed at /v2/search/developer and via the firecrawl_developer_search MCP tool. TinyFish returns fresh live web results but does not maintain pre-built specialty indexes for these domains, so agents doing scientific research or answering questions from library changelogs get more relevant sources with fewer hops on Firecrawl.

Most TinyFish calls map cleanly to Firecrawl. TinyFish Search maps to Firecrawl /search, TinyFish Fetch maps to Firecrawl /scrape (with actions and JSON Schema now available on the same call), TinyFish Agent maps to Firecrawl /agent, TinyFish Research maps to Firecrawl's research modes on /agent, and TinyFish Browser maps to Firecrawl /browse. Swap the base URL and API key, adjust the request shape, and reuse the same response fields. Teams that also need full-site coverage add /crawl, teams that need file parsing add /parse, and teams that need to watch pages for changes add /monitor for scheduled diffs and webhooks. Firecrawl SDKs cover Python, Node.js, Go, Rust, Java, .NET, Ruby, PHP, and Elixir, and most migrations take a day or less.

Yes. Firecrawl is SOC 2 Type II compliant with GDPR compliance and a DPA available. Enterprise plans include zero data retention and a 99.9% SLA. Self-hosting under AGPL-3.0 is available for air-gapped environments, and the managed cloud has served more than 5 billion requests to date for over 150,000 companies. TinyFish also offers an enterprise plan with ISO 27001, SSO, audit logs, and VPC deployment. Both are viable at enterprise scale. Firecrawl's open source and self-hostable platform is the differentiator for teams that need to audit, fork, or fully own the codebase.

Yes. Firecrawl Alexandria is a library of 90+ data providers that Firecrawl returns as tools inside a search call: add alexandria to the sources parameter and the response carries a tools array next to the web results, covering official APIs such as SEC EDGAR, FRED, and the World Bank, licensed publishers such as Fiscal.ai and Benzinga, connectors such as the Wayback Machine and Greenhouse job boards, and Firecrawl's Research, Developer, and Government indexes. Finding a tool is free and running one is billed at the price the tool lists. TinyFish's Agent and Browser run against websites, so a structured provider is a separate integration. With Firecrawl the same API key and search call reach both the web and those providers.