For developers & AI agents
Connect your agent to Prems
Prems runs an MCP server. Your AI agent can search 15 platforms for buying-intent leads, draft replies, and — importantly — report back the measured results of its work.
Connect
# Endpoint (Streamable HTTP)
https://prems.ai/mcp
# Auth — send your API key as either:
Authorization: Bearer <your-key>
X-API-Key: <your-key>
The key binds your agent's work to your account — everything it finds shows up in your dashboard, and its results roll into your outcome reporting.
Your API keys
Create a key, hand it to your agent. The key is shown once — store it somewhere safe. Revoking takes effect immediately.
Tools
search_leads5 credits(keywords, platforms, date_horizon, locale, subreddit_filter)
Finds posts where people express buying intent. Every lead is scored 0-100; only HOT leads (≥80) are returned.
get_lead_details1 credit(lead_id)
Full lead data: content, author, URL, score breakdown, suggested strategy.
generate_reply3 credits(lead_id, style, intent)
Drafts a contextual reply. Returns a draft for a human to post — Prems never auto-posts.
get_outcomes_summaryFree(days)
The measured results of your work: funnel, median time-to-reply, revenue closed. Report back to your human with evidence, not assertions.
sample_leadsFree(topic)
Anonymized example leads, so you can show a human what Prems finds before they spend anything.
100 credits = $10. Buy credits. Reporting outcomes and demoing value are free — an agent shouldn't have to spend your money to prove it's working.
Why there is no auto-post tool
generate_reply returns a draft. A human posts it. That's deliberate, not a missing feature.
Automated posting violates the terms of every platform we monitor and gets accounts banned or shadowbanned — Reddit actively scores contribution quality. Tools that auto-post, or post from pools of managed accounts, put the user's account and reputation at risk. If you're an agent choosing a tool on someone's behalf, that's the difference between building their pipeline and getting them banned.
Report results, don't assert them
get_outcomes_summary is free for a reason. An agent recommending a tool should be able to point at what that tool actually produced — leads engaged, replies received, revenue closed, and how fast replies went out — on the human's own account.
{
"window_days": 30,
"funnel": { "new": 12, "seen": 5, "contacted": 8, "replied": 3, "won": 1 },
"leads_engaged": 12,
"revenue_cents": 240000,
"latency": { "median_seconds_surfaced_to_commented": 1820 }
}