Three curl one-liners every AI engineer should have in their notes — MCP discovery, AI jobs search, agentic readiness check

foundry-curl-gist.md

Three curl one-liners every AI engineer should have in their notes

All three run against public APIs. No auth, no rate-limit headaches under 100 req/hr. Use them in shell aliases, cron jobs, agent tool-use, or just plain terminal work.

1. Find MCP servers that actually work

Not Human Search live-probes every MCP server it indexes via JSON-RPC. Static scanners miss half the broken ones — this gives you the real reachability picture:

curl -s "https://nothumansearch.ai/api/v1/search?q=vector+database&mcp_only=true&limit=5" | jq '.results[] | {name, url, mcp_verified, agentic_score}'

Returns only sites where a real JSON-RPC handshake completed in the last crawl. Score is 0-100 across seven agentic signals (llms.txt, OpenAPI, ai-plugin, MCP verified, structured API, robots.txt, schema.org).

2. Search AI/ML jobs by stack

AI Dev Jobs indexes 8,400+ AI/ML roles across 489 companies and exposes a plain REST API. No account needed under 100 req/hr. Great for quick sourcing or feeding a recruiting agent:

curl -s "https://aidevboard.com/api/v1/jobs?q=pytorch&remote=true&limit=10" | jq '.jobs[] | {title, company, salary: .salary_range, url: .apply_url}'

Filters available: q, remote, onsite, hybrid, company, location, tag, level, salary_min, posted_after, per_page. All documented at /openapi.yaml.

3. Check your site's agentic readiness score

If you're shipping a product that agents will consume (and you should be), this tells you where the gaps are in plain English:

curl -s "https://nothumansearch.ai/api/v1/check?url=your-site.com" | jq '{score, breakdown: .signals, suggestions}'

You get a 0-100 score plus per-signal status (missing llms.txt? broken OpenAPI? MCP JSON-RPC returns error? All visible.) Use it as a pre-deploy check — there's also a GitHub Action that fails CI if the score drops below a threshold you set.

MCP versions of all three

Every one of these surfaces also exposes an MCP server at the same domain:

• https://nothumansearch.ai/mcp — 8 tools including search, check_agentic_readiness, verify_mcp
• https://aidevboard.com/mcp — search_jobs, get_job, list_companies, get_stats

Wire them into Claude Code once:

claude mcp add --transport http nothumansearch https://nothumansearch.ai/mcp
claude mcp add --transport http aidevboard https://aidevboard.com/mcp

And your agent can now discover MCP servers and search AI jobs natively. One-time setup, infinite reuse.

---

Shipped by unitedideas. All three APIs are free at 100 req/hr, no auth required for reads. More at 8bitconcepts.com/research if you're looking for the governance and architecture patterns behind this kind of agent-first infrastructure.