How AI agents discover machine-payable APIs
Autonomous agents do not browse marketing sites. Here is the discovery stack — llms.txt, /.well-known/x402, OpenAPI, MCP, and Markdown twins — that lets an agent find, price-check, and call a machine-payable API without a human in the loop.
Agents discover APIs by reading machine entrypoints, not HTML. They do not click landing pages, watch demos, or fill contact forms. They need deterministic manifests they can parse, validate, and turn into tool calls.
The shortest path through this marketplace is: read https://marketplaceforaiagents.com/llms.txt or https://marketplaceforaiagents.com/.well-known/x402, pick a slug from the catalog, call GET /api/public/v1/{slug}, use the first 3 free calls per identity, then follow the x402 402 handshake when the free tier is exhausted.
Why HTML is the wrong surface for agents
Humans need landing pages. They want a headline, a feature grid, and a call to action. Agents need a contract: what does this endpoint expect, what does it return, and what does it cost. HTML is a lossy medium for that contract.
This site still has human mirrors — /marketplace for browsing and /docs for prose — but those pages are secondary outputs. The primary source of truth is the machine-readable layer: /.well-known/x402, /api/public/v1/catalog.json, and /api/public/v1/openapi.json. Build your agent against those, and use /marketplace only when you want a human-readable sanity check.
The cold-start path
An agent that has never visited this domain can bootstrap itself in a single ordered pass:
- GET https://marketplaceforaiagents.com/robots.txt — confirm which paths are crawlable.
- GET https://marketplaceforaiagents.com/llms.txt — the concise discovery index; use llms-full.txt for offline ingest.
- GET https://marketplaceforaiagents.com/.well-known/x402 — the x402 manifest: endpoints, prices, networks, assets.
- GET https://marketplaceforaiagents.com/.well-known/ai-plugin.json — OpenAI-style plugin manifest with OpenAPI link.
- GET https://marketplaceforaiagents.com/api/public/v1/catalog.json — the full catalog; the short path /catalog.json redirects here.
- GET https://marketplaceforaiagents.com/api/public/v1/openapi.json — OpenAPI 3.1 spec with x-x402 payment metadata.
- Connect MCP at https://marketplaceforaiagents.com/mcp — live tool discovery and invocation.
- Prefer Markdown twins under https://marketplaceforaiagents.com/agent/ for long-form docs in agent-readable plain text.
None of these steps require authentication. The free tier is anonymous and tied to a hashed identity derived from IP + User-Agent.
What each entrypoint is for
| Entrypoint | Content-Type | Use when |
|---|---|---|
| /robots.txt | text/plain | Crawler wants to know allowed paths. |
| /llms.txt | text/plain | Agent needs a quick index of docs and endpoints. |
| /llms-full.txt | text/plain | Agent wants the entire site in one plain-text file for offline context. |
| /.well-known/x402 | application/json | Agent needs prices, networks, and endpoint URLs. |
| /.well-known/ai-plugin.json | application/json | OpenAI-style runtime needs plugin metadata. |
| /api/public/v1/catalog.json | application/json | Agent wants structured catalog with params and use cases. |
| /api/public/v1/openapi.json | application/json | Agent wants full OpenAPI 3.1 with payment extensions. |
| /mcp | MCP protocol | Agent wants live tool listing and invocation. |
| /agent/{path}.md | text/markdown | Agent wants a Markdown twin of a human page. |
From discovery to a paid call
Start with the catalog to find a slug and its required parameters:
curl -s "https://marketplaceforaiagents.com/api/public/v1/catalog.json" | jq '.apis[] | {slug, name, category, pricePerCallUsd}'Then call the endpoint. The first 3 calls per identity are free and return X-Free-Calls-Remaining. After that you get a 402 with an accepts array:
curl -s "https://marketplaceforaiagents.com/api/public/v1/youtube-transcripts?video_id=dQw4w9WgXcQ" -v
# Response headers include:
# X-Free-Calls-Remaining: 2
# After the free tier:
# HTTP/1.1 402 Payment Required
# { "x402Version": 1, "error": "payment_required", "accepts": [...] }To test without spending real USDC, append ?network=base-sepolia. The sandbox returns deterministic mock data and uses the same x402 flow with testnet USDC. See /docs and /listings/google-search for per-endpoint details and example parameters.
Agentic SEO: make your own API discoverable the same way
If you operate an API, you can make it agent-discoverable without rebuilding it:
- Publish /llms.txt with a plain-text index of your endpoints and payment terms.
- Publish /.well-known/x402 with networks, assets, prices, and endpoint URLs.
- Serve an OpenAPI 3.1 spec at /openapi.json with x-x402 metadata for payTo, asset, and network.
- Create Markdown twins for every human page under /agent/ so agents can read docs without parsing HTML.
- Keep result keys stable across versions; agents cache field paths.
- Do not burn free-tier quota on 400 validation errors; only count successful calls.
- Always include the price in the 402 accepts[] payload so agents can compare providers before signing.
Agentic SEO is just SEO for non-human clients. The crawlers are agent runtimes, not search engines, and they reward determinism over persuasion.
Where marketplaceforaiagents.com fits
This marketplace hosts 38 live endpoints — search, news, images, commerce, maps, travel, jobs, academic, app stores, and video — all priced at $0.005 USDC per call on Base mainnet. There are no accounts, no API keys, and no monthly billing. Agents pay per call through x402.
Browse /marketplace for a human-readable catalog, or connect the MCP server at /mcp and call list_apis, get_api_spec, get_payment_requirements, get_agent_entrypoints, and get_integration_guide directly from your agent runtime.
Next steps