Search Web

Keenable Keyless Web Search

Two endpoints on https://api.keenable.ai, both usable with no key and no account:

Both require the header X-Keenable-Title, whose value is any non-empty string identifying the calling application - the examples below use agent-skills. It has no default; omitting it returns 400 Missing app identifier.

Search

curl -s -X POST "https://api.keenable.ai/v1/search/public" \
  -H "X-Keenable-Title: agent-skills" \
  -H "Content-Type: application/json" \
  -d '{"query":"model context protocol spec"}'

Body parameters. query is the only required one; the API rejects unknown parameters with 400, so send only names from this table:

The two bounded ranges are enforced by the server but absent from its OpenAPI schema. max_results: 100 and snippet_max_length: 150 both return 400 Invalid parameter. Stay inside the bounds above rather than trusting the published schema.

Response:

{
  "query": "model context protocol spec",
  "mode": "pro",
  "results": [
    {"title": "...", "url": "https://...", "description": "...", "snippet": "...",
     "published_at": "2026-05-18", "acquired_at": "2026-08-25"}
  ]
}

Only title, url, and description are guaranteed present on a result. Treat snippet, published_at, and acquired_at as optional and check before reading them.

Fetch

curl -s -G "https://api.keenable.ai/v1/fetch/public" \
  --data-urlencode "url=https://en.wikipedia.org/w/index.php?title=Markdown" \
  -H "X-Keenable-Title: agent-skills"

Use -G with --data-urlencode so the URL is escaped correctly; a raw ?url= breaks on query strings and fragments.

Response: {"url", "title", "content", "description", "author", "published_at"}, where content is markdown. Only url and content are guaranteed. url reflects the final address after redirects, so it may differ from what was requested.

Default to the indexed copy. Set live=true only when the page must be current as of right now, since it is slower and hits the origin server.

Limits and errors

Keyless quota is 1,000 requests per hour and 10 per second, counted per IP — shared with everything else behind the same egress address, so the available headroom is not under this machine's control. No credits are consumed.

Every response carries x-ratelimit-limit, x-ratelimit-remaining, and x-ratelimit-reset (an ISO 8601 timestamp). Read x-ratelimit-remaining before starting a batch of calls.

Error bodies are {"error": "...", "message": "..."}. On a 400, message states the exact constraint — parse it instead of guessing.

Never retry a 400 with the same body. It is deterministic and will fail identically.

Scope

Use this skill for the open web: current events, release notes, changelogs, blog posts, standards, project status, anything past the training cutoff.

Do not reach for it when a better-targeted tool exists:

• Repository contents → grep and glob, not web search.
• Library and framework API reference → context7, which returns version-pinned docs. Two exceptions come back here: material context7 does not carry, such as a changelog or an announcement post; and any request that names a site, a date window, or a specific URL, since context7 cannot scope a search that way.
• General concepts, algorithms, and design questions resting on no external fact → answer directly.

research-it is not on that list. It owns investigation method — source priority, triangulation, conflict reporting — and calls this skill to do the actual retrieval. When a task needs both, run both.

Notes

The MCP server at https://api.keenable.ai/mcp also answers keyless, exposing search_web_pages and fetch_page_content. Prefer the HTTP endpoints above when running one-off calls from a shell; they need no client configuration.

Omit the mode parameter. The keyless endpoint accepts pro and realtime and echoes the choice back, contradicting Keenable's docs on two counts: that mode is not a request parameter, and that realtime needs a key. Undocumented behaviour that the vendor's own docs deny can be withdrawn without notice. Keyless calls default to pro.