TinyFish opens Search and Fetch for free with MCP, CLI, and <0.5 s p50

• TinyFish opened its web testingcatalog launch thread and TinyFish blog post around Search and Fetch for free, with no credit card required and free-tier rate limits listed on TinyFish pricing.
• According to ai_for_success's summary, Search returns structured JSON for agent retrieval, while Fetch docs say Fetch renders JavaScript-heavy pages and returns clean markdown, JSON, or HTML.
• The rollout is broad: testingcatalog's follow-up says the same API key works across REST, MCP, CLI, Skills, and Python and TypeScript SDKs, while MCP integration docs show quick installs for Claude Code, Claude Desktop, Cursor, and Windsurf.
• Free does not mean the whole platform is free. TinyFish pricing says Search and Fetch cost nothing on every plan, but agents and browser sessions still consume credits, starting with 500 credits on the free plan.

You can read the official announcement, skim the pricing page, and jump straight into the MCP install guide. The docs also split the product cleanly into Search, Fetch, Agent, and Browser, which makes the pricing change more interesting than the tweet version suggests.

The headline is simple: Search and Fetch are now free for every developer and agent, and TinyFish is framing them as production-usable from day one. testingcatalog's launch thread says the company removed the card gate, while TinyFish pricing lists the free plan at $0 per month with 500 starter credits.

The more concrete detail lives on pricing, not in the announcement copy. TinyFish pricing lists free-tier limits of 5 search requests per minute and 25 fetched URLs per minute, and the Fetch API reference shows that Fetch requests can include up to 10 URLs at a time.

TinyFish is separating two jobs that agent builders usually jam together.

• Search docs describe Search as a structured results API that returns titles, snippets, and URLs, not a search page meant for humans.
• Fetch docs describe Fetch as a real-browser renderer for pages you already know you want, with output in markdown, JSON, or HTML.
• The Fetch API reference marks markdown as the default and recommended format for LLMs.
• The same reference says private IPs, localhost, and cloud metadata endpoints are rejected.

That split matters because TinyFish is making the retrieval layer free, not the full automation stack. testingcatalog's follow-up pitches Fetch as a way to avoid feeding raw HTML to models, and ai_for_success's summary adds a p50 under 0.5 seconds for Search, which is fast enough to sit inside a tool loop if the number holds up in broader use.

The distribution story is almost the real launch. testingcatalog's follow-up says Search and Fetch are available through REST, MCP, CLI, Skills, and Python and TypeScript SDKs, plus integrations for Claude Code, Cursor, Codex, n8n, Dify, LangChain, CrewAI, and Vercel Skills.

The official docs show how much of that is already wired up. MCP integration docs include one-command installs for Claude Code, Claude Desktop, Cursor, and Windsurf, plus a manual claude mcp add --transport http tinyfish https://agent.tinyfish.ai/mcp path for Claude Code. The CLI docs also show a separate terminal workflow that stores credentials in ~/.tinyfish/config.json.

The pricing page draws a hard line between retrieval and execution. TinyFish pricing says Search and Fetch are free on every plan, but credits still apply when you run agents or browser sessions.

That shows up in the product map too. The developer docs home lists four public surfaces, Agent, Search, Fetch, and Browser, and the free change only covers two of them. testingcatalog's linked article makes the same point indirectly by describing Search and Fetch as the low-friction entry point, with the same key and dashboard carrying usage into paid tiers once teams move beyond retrieval.