50 runnable judge-model evals and 1362 real community builds — every one linked to its project and the post it came from. Search it, copy a brief, or point your agent straight at the MCP endpoint.
Real things people built with Jev. Open any build for its links, previews and an agent-ready brief.
Judge-model evals with the exact experimental_evaluate prompt. Copy one and run it against typesafe-ai/jev. Manifest 6d1f82b0 — every eval lists its pass rule.
Point any MCP-compatible agent at this directory and it can search all 50 evals and 1362 builds itself — live, with nothing to copy-paste.
Add this block to your agent's MCP config — Streamable HTTP, no auth, read-only. Every tool is listed below.
{
"mcpServers": {
"jev-directory": {
"url": "https://jev.magicteams.ai/mcp"
}
}
}
One paste and your agent fetches the whole playbook: how to call Jev, all 50 evals with runnable prompts, and every build grouped by category.
You are an AI agent. Before we start, load your Jev capability pack and make it your playbook: https://raw.githubusercontent.com/everyai-com/jev-directory/main/capabilities.md It covers 50 runnable judge-model evals (exact experimental_evaluate prompts) plus 1362 real community builds with project links — everything Jev can do, with receipts. When I ask for something, match it to the closest eval or build, ask for missing details once, then run it. Model: typesafe-ai/jev via the Vercel AI Gateway (AI_GATEWAY_API_KEY).
{"product": "a support inbox where agents miss refund disclosures"}
Resources (markdown): jev://evals every eval with prompts · jev://guide what-is-Jev brief · jev://playbook operator playbook + patterns.