Skip to content
MCP Atlas

Developers & AI agents

MCP Atlas is machine-readable by design: server-rendered HTML, structured data, a sitemap, llms.txt, and a public read-only JSON API. No account or API key is required.

Public JSON API (v1)

Base URL: this site's origin. All endpoints are GET-only, return JSON, and require no authentication. An OpenAPI 3.1 specification is available at /openapi.json.

  • GET /api/v1/servers — paginated list. Parameters: q, category, transport (stdio|http), hosting (self-hosted|cloud), verified (true), sort (updated|stars|verified|name), page (1–200), per_page (1–50, default 20).
  • GET /api/v1/servers/{slug} — one server with provenance and publisher profile. 404 with a JSON error body when the slug does not exist.
  • GET /api/v1/categories — categories with live server counts.
  • GET /api/v1/sources — data sources with ingestion mode and server counts.
  • GET /api/v1/search?q=... — search; q is required, per_page is capped at 25.
curl -s "https://zeroo.online/api/v1/servers?category=databases&sort=stars&per_page=5"
curl -s "https://zeroo.online/api/v1/search?q=postgres"
curl -s "https://zeroo.online/api/v1/servers/modelcontextprotocol--filesystem"

If /api/v1/* is unreachable from your network, the identical endpoints are mirrored under /api/public/v1/*, which is guaranteed to bypass site-level access control.

Rate limits, caching and CORS

  • Rate limit: 60 requests per minute per client IP, enforced best-effort per edge isolate. Exceeding it returns 429 with a Retry-After header — honour it.
  • Every response carries an ETag and Cache-Control. Send If-None-Match to receive a cheap 304 Not Modified. List and detail responses also carry Last-Modified.
  • CORS is enabled for public GET requests from any origin (Access-Control-Allow-Origin: *). Non-GET methods are not supported.
  • Query complexity is bounded: short query strings, capped page sizes, and a maximum page depth of 200. Please crawl politely and cache responses.

Discovery files for AI agents

  • /llms.txt — short plain-text site description with links.
  • /llms-full.txt — extended context: data-quality rules, categories, API usage, attribution requirements.
  • /sitemap.xml — canonical URLs for every live listing and content page.
  • /robots.txt — all crawlers are welcome except under /admin. Note that robots.txt rules are voluntary: they express our policy but do not guarantee that any agent will index — or respect — the site.

All listing and detail pages are server-rendered: titles, descriptions, categories, publisher, source attribution and last-updated times are present in the HTML without JavaScript, and each page carries JSON-LD structured data (WebSite, Organization, SoftwareApplication, BreadcrumbList).

Connect your agent (public MCP server)

Endpoint: https://zeroo.online/mcp — a remote, read-only MCP server over Streamable HTTP (stateless, POST only). No login or API key. Verified externally with the official MCP SDK client on 21 October 2026 (initialize, tools/list and tools/call for all six tools).

  • search_servers — keyword, category, transport, license, minimum known stars, pagination.
  • get_server_details — one listing by slug with provenance, publisher and evidence-backed compatibility.
  • list_categories — categories with live counts.
  • list_recent — newest by real registry update time.
  • compare_servers — 2–4 slugs side by side.
  • recommend_servers — deterministic, explainable scoring with confidence and reasons. Not an endorsement.
{
  "mcpServers": {
    "mcp-atlas": { "type": "http", "url": "https://zeroo.online/mcp" }
  }
}
claude mcp add --transport http mcp-atlas https://zeroo.online/mcp

Limits: 120 requests/minute per client IP (best-effort, 429 with Retry-After), 32 KB request body, 25 results per page. Browser calls from other websites are refused (403). The tools only read public directory data — no writes, installs, code execution, URL fetching, submissions or admin data.

What this API is not

The website pages are not an MCP transport; only /mcp is. The MCP Atlas discovery server is separate from any AiWorkPay business MCP. The planned first-party AiWorkPay MCP endpoint (https://mcp.aiworkpay.fr/mcp) is labelled "Planned / Not verified" — do not treat it as live.

The API exposes only public catalog data. It never returns user data, secrets, admin endpoints, or unpublished repository content. Write operations (submissions, synchronization) are not part of the public API.

Attribution and fair use

Listings aggregate third-party metadata under the terms of their source APIs (see each listing's provenance and the sources page). When citing MCP Atlas data, attribute both MCP Atlas and the underlying source. Unknown values are null — never treat a null as zero or as a negative claim.