For agents

Search and cite a directory of 3755 graded typed-decision AI use cases. Read access needs no key or account. OpenAPI describes the public endpoints; full dataset and taxonomy are downloadable.

Send a descriptive User-Agent (e.g. Hermes-Agent/1.0 (+contact)); some generic library defaults such as Python-urllib are blocked by Cloudflare's browser integrity check.

API

GET /api/use-cases accepts q, grade, vertical, function, status, limit (maximum 200), and offset. GET /api/use-cases/{slug} fetches one record. GET /api/health checks service health.

curl 'https://jevaiusecases.com/api/use-cases?q=triage&grade=S&limit=10'
curl 'https://jevaiusecases.com/api/use-cases/prompt-injection-detection-guardrail'

MCP clients

The remote Streamable HTTP MCP server is https://jevaiusecases.com/mcp. It exposes search_use_cases, get_use_case, get_submission_challenge, and submit_use_case.

• **Claude Desktop:** add a remote MCP server URL https://jevaiusecases.com/mcp in its connector settings.
• **Claude Code:** add a remote HTTP MCP server with URL https://jevaiusecases.com/mcp using your installed version's MCP command or settings.
• **Hermes:** add remote MCP server URL https://jevaiusecases.com/mcp to its MCP configuration.
• **OpenClaw:** add remote MCP server URL https://jevaiusecases.com/mcp to its MCP configuration.

Submit from an agent

New use cases enter a moderated queue; none publish automatically. Required fields: title (8–120 chars), description (40–1000), taxonomy vertical/function, and decision_type (yes_no, choice, score). Optional: example_input, source_url, submitter_name, submitter_contact, agent_name (your agent's name, e.g. Hermes or OpenClaw). Keys are listed at /api/taxonomy.json. Leave website empty. Fetch GET /api/challenge, solve SHA-256 of challenge:nonce for difficulty leading zero bits, and send pow: {challenge, nonce} with POST /api/submissions. A challenge expires in 10 minutes and is single use. Rate limits: 5/hour and 20/day per IP hash. Humans may use Turnstile instead. Every submission is moderated.

import hashlib, json, urllib.request
base = 'https://jevaiusecases.com'
c = json.load(urllib.request.urlopen(base + '/api/challenge'))
n = 0
while True:
    h = hashlib.sha256(f"{c['challenge']}:{n}".encode()).digest()
    if int.from_bytes(h, 'big') >> (256 - c['difficulty']) == 0:
        break
    n += 1
body = {'title': 'Classify urgent repair requests', 'description': 'Decide whether each incoming property repair request needs immediate human escalation based on safety and severity.', 'vertical': 'real_estate', 'function': 'support_cx', 'decision_type': 'yes_no', 'agent_name': 'MyAgent', 'website': '', 'pow': {'challenge': c['challenge'], 'nonce': str(n)}}
req = urllib.request.Request(base + '/api/submissions', data=json.dumps(body).encode(), headers={'Content-Type': 'application/json'}, method='POST')
print(urllib.request.urlopen(req).read().decode())

Citation and attribution

Link each cited item to its canonical /use-cases/{slug}/ page. Name its grade and any source URL. Grades express editorial judgment, not operational approval. You may add this badge to a site using the directory:

<a href="https://jevaiusecases.com/" rel="noopener">Powered by Jev AI Use Cases</a>