Web search and content retrieval have become critical infrastructure for AI agent development. Agents lacking reliable access to live web data operate on stale knowledge, posing a significant limitation for production deployments requiring real-time information for tasks like research, lead enrichment, competitive intelligence, or monitoring. By 2026, the ecosystem of search and fetch APIs has matured, offering purpose-built tools that surpass older methods of directly feeding raw Google SERP data into language models.
Among the leading search and fetch APIs, TinyFish is a notable entrant, distinguished by its directly agent-native design. Both its Search and Fetch endpoints are free, offering generous rate limits: 5 requests per minute for Search and 25 requests per minute for Fetch, accessible with one API key and no credit card requirement.
TinyFish's Search API, operating at api.search.tinyfish.ai, delivers rank-stable structured JSON output, specifically optimized for agent retrieval rather than human browsing. It boasts a p50 Search latency of under 0.5 seconds, fast enough to integrate within an agent’s tool loop without degrading user experience.
The Fetch API, found at api.fetch.tinyfish.ai, performs a full-browser render on any URL. This includes complex pages like JavaScript-heavy Single Page Applications (SPAs), dynamically loaded content, and anti-bot protected sites. It returns clean markdown, JSON, or HTML, and failed URL attempts are not charged.
A key differentiator for TinyFish is its token efficiency. Unlike many native fetch tools or LLM client-integrated fetch functions that return raw HTML (including scripts, navigation, ads, and cookie banners), TinyFish Fetch preprocesses content. It strips away all non-essential elements before the content reaches the language model, significantly reducing token consumption per page and lowering LLM costs per call.
The platform runs its own custom Chromium fleet end-to-end, without middleware. This proprietary infrastructure enables both its free pricing model and high output quality. These are production-grade endpoints, not a scaled-down demo tier, ensuring seamless transition and code compatibility when scaling beyond the free plan.
TinyFish provides extensive integration options for developers. Direct access is available via REST API (api.search.tinyfish.ai and api.fetch.tinyfish.ai). For Multi-Client Protocol (MCP) compatible environments, a single JSON configuration drop-in supports clients like Claude, Cursor, Codex, ChatGPT desktop, and other MCP-aware platforms.
A Command Line Interface (CLI), installed via npm install -g @tiny-fish/cli, writes results directly to the filesystem, bypassing the model’s context window to further minimize token usage and maintain structured output. Furthermore, an agent Skill (npx skills add github.com/tinyfish-io/tinyfish-cookbook –skill tinyfish) teaches agents when and how to utilize the Search versus Fetch capabilities, compatible with tools such as Claude Code, Codex, Cursor, OpenCode, and Antigravity. Python and TypeScript SDKs are also provided.