The eight we hear most.
What is MCP, and why does my agent need it?
MCP (Model Context Protocol) is an open standard for AI agents to call external tools. Claude, ChatGPT, Cursor, Codex, Continue, Cline, Zed, and Windsurf all speak it natively. AgenticSource is one MCP server exposing 15 media tools, so your agent connects once and can fetch any kind of media without juggling 27 different upstream APIs.
How is this different from calling Unsplash, Pexels, and Iconify directly?
AgenticSource is the agent-first version of those APIs. Upstream sources were designed for humans clicking through dashboards, not agents reasoning in real time. Direct integration means your agent juggles 27 different response shapes, 27 license formats, and 27 rate limits across 11 categories, with zero dedup on overlapping results. AgenticSource normalizes all of it into one MCP tool surface: one connection, one response shape, one license model. Cross-source ranking, dedup, KV caching, and graceful upstream failover are already shipped. Your agent reasons about media once instead of 27 times.
What does "license-aware" actually mean?
Every result carries four fields: license_name, license_url, commercial_use (boolean), and attribution_required (boolean). Pass commercial_safe_only: true on any search and we filter out anything your agent shouldn't ship. The point is that an agent can reason about usage rights instead of guessing.
Can I use the results commercially?
Yes. Every result passes through its upstream license unchanged (CC0, CC-BY, Unsplash License, Pexels License, etc.), and your agent gets four metadata fields on every asset: license_name, license_url, commercial_use, and attribution_required. Results tagged commercial_use: false (mostly CC-BY-NC content) are filtered out automatically when you pass commercial_safe_only: true, which is the default for ad and product use cases. We do not generate AI imagery — every asset is real, licensed, and traceable to its source.
How fast is it?
Median 400ms per search on cold paths. KV cache hit rate sits at 70 to 80% in steady state, so most repeat queries return in under 50ms.
What happens when I hit my plan limit?
On Free, search tools return a 429 with a friendly message until your monthly counter resets. On Pro and Stock+, you stay productive: hit your cap and either auto-buy a boost pack (if you've enabled it) or get a one-click "Buy Boost" button on the next call. No silent fails, no surprise upgrades.
What if an upstream source goes down or rate-limits us?
Multi-source categories degrade gracefully. If Unsplash is unreachable, photo searches still return results from Pexels, Pixabay, OpenVerse, Wikimedia, NASA, and Met Museum. The response shape stays identical; a sources_responded array tells your agent who answered. Single-source categories (Logos via Logo.dev, 3D via Poly Haven) have less redundancy: if they're down, the tool returns a structured error your agent can fall back from.
Can I self-host?
Yes. The server is MIT-licensed and runs on Cloudflare Workers, D1, and KV. Full source on GitHub. You bring your own upstream keys for Unsplash, Pexels, Pixabay, Freesound, etc. Most have generous free tiers; a few are free for non-commercial use only.
