The strike sequence — drag
Six cuts.
No wasted motion.
I. Compile
Your sentence becomes a battle plan — geography, industry, size, signals, testable criteria. Business context sharpens buyer-fit.
II. Discover
Four search backends rotate so no index owns the pool. 2.5× candidates; research waves loop until the count fills.
III. Research
Robots-gated fetches, cached pages, embedded passages. Every PASS cites the exact lines that earned it.
IV. Qualify
Code-graded PASS / FAIL / UNKNOWN. Conflicts and rot become UNCERTAIN — never averaged into false confidence.
V. Deliver
Qualified plus capped uncertain ship with per-lead reasons. The disqualified are listed with their wounds, never counted.
VI. Refresh
Reruns reuse fresh evidence and re-cut stale pages — REFRESHED, FRESH, STALE, or GONE, each explained.
Verdicts, explained
Every state
has a reason.
2/2 criteria pass — location matched, industry matched, both cited to retrieved passages.
No hiring evidence, no size evidence — declared unknown instead of invented. Needs review.
Content about targets is not a target — page is an article. Listed with reasons, never counted.
Battle record
Proven live.
MCP tools
Routed sources
Checks green
Output formats
The arsenal — MCP
One blade.
Every hand.
REAVER is an MCP server. Your AI host does the thinking; REAVER supplies ten lead-intelligence weapons — prospect runs the full hunt end-to-end, nine specialists handle discovery, research, enrichment, qualification, verification, dedup, refresh, and export.
Local Claude / OpenCode
One line, localhost, no auth needed:
claude mcp add reaver -- python -m reaver_mcp.server --transport stdio
Remote ChatGPT / Claude API
Expose over HTTPS with a bearer token, then paste the URL as an MCP server. Token enforced, localhost stays frictionless.
Same Playground
This site's Playground drives the identical tools with your own keys — no second implementation, no drift.
One weapon.
Every arena.
The same MCP serves ChatGPT, Claude, OpenCode — and this Playground, on your keys and your models.