Evergreen turns a plain-language acquisition or sales thesis into an investment-grade list — searched from the Evergreen Index first, deduplicated, enriched, thesis-scored, and QA’d — in Claude, Codex, Cursor or any MCP client. Powered by the Evergreen API.
evergreen❯ commercial landscaping in Texas, owner-operated, no PE ● evergreen - start (MCP) ⎜ session opened · playbook: screen ⎜ Q1 size band? Q2 ownership? Q3 end market? ⎜ (answer choices from your org's schemas) ❯ 20–200 staff · founder or family owned · commercial only ● evergreen - find_companies (MCP) ⎜ 12 phrasings × 6 metros over the Evergreen Index ⎜ ~1,000 distinct companies · exact market count alongside ● evergreen - stage_run (MCP) intake → screen → score ⎜ deduped · blocklist + ledger applied · ranked ● Add Google Maps + Exa for metros the Index covers thinly? ❯ not now ● evergreen - stage_run (MCP) export # qualified targets — ranked, with reasons and provenance download: commercial-landscaping-tx.xlsx
What it does
Where research answers questions about one company, Evergreen builds and qualifies the entire universe of companies that match a thesis.
Every search starts in the Evergreen Index — 0.1 credit per company, once per session, and fast: many phrasings across every location at once, with an exact count of the market beside what was reached.
Size, ownership, end market and location questions come from your own schemas, so the answers become the filter plan. A missing condition gets a proposed schema instead of a guess.
Google Maps POI, Exa neural search and Ocean.io lookalikes run only when you approve them — then fan out across metros and queries and dedupe by registrable domain.
Paid research runs as a 50–100 company pilot you review first. The full pool waits until accuracy clears 90% — no batch of credits is spent on a schema that has not proven itself.
Mandatory ownership re-verification flags PE roll-ups that look independent; reserved coverage probes catch what the main run missed. Confirmed roll-up brands feed a curation queue for future runs.
Each run is a server-side session: every tool call, answer, gate and saved stage is recorded, so a run resumes from any client and the whole journey is auditable in the platform.
Describe the target profile in a sentence — or attach a spreadsheet, PDF or document of companies and ask a question about them. No commands to learn.
One connector and nothing to install. Claude chat, Claude Code, Codex, Cursor and Copilot all get the same tools and the same full pipeline.
Pipeline
Every run is a server-side session — every tool call is recorded, and any client picks up exactly where the last one stopped. Credit gates sit between stages, paid discovery waits for your consent, and the pipeline stops for a human before money moves and again before anyone is contacted.
The ask is read before anything is searched: the scenario, the trade in several phrasings, the geography expanded (“Sun Belt” becomes its states), and the few questions that decide the list — size, ownership, end market — with answer choices drawn from your organization’s schemas. Where no schema covers a condition, one is proposed. You also name two or three companies that must appear, so the run can show it found them.
When there is a buyer to research — a PE backer, a platform to bolt onto — the firm and its portfolio are studied first, producing a thesis and a buy-box that seed the plan. Skipped when the request is self-contained.
The Evergreen Index is searched before anything is bought: many phrasings of the trade across every location in one call, unioned, with an exact market count beside what the search reached. 0.1 credit per company, charged once per session, and usually seconds.
Only if you approve it: live discovery through Google Maps (POI), Exa (neural) and Ocean.io lookalikes, cross-source deduped by registrable domain with website resolution and per-source provenance. The server refuses a List Building call that no one approved, and the PE/franchise blocklist is applied at discovery.
Server-side screening before a credit is spent: dedupe, your project blocklist, the client’s conflict ledger, and website-gap flags. A ranked preview lands here too, so a simple ask can stop at zero spend.
Free page evidence first, then re-extraction of research already paid for, then paid research on survivors only. Tier 1 runs as a 50–100 company pilot behind a review gate — the full pool does not run until accuracy clears 90% or you explicitly override.
Hard requirements first, then a scored checklist on the survivors — every company carries the reason it passed and what is still unverified. Ranked, not just listed.
An independent judge on the server spot-checks finalists against the evidence, re-verifies ownership on every sampled company (catches PE roll-up brands that keep local names), and runs reserved coverage probes for misses.
Website liveness, HQ and legal name are verified, and export blocks while any issue is unresolved — a deliverable never ships with a dead link or a wrong HQ. You get an XLSX or CSV download link.
The human approval gate. Conflict checks, territory rules, a partner’s eyeball — whatever your review is, the pipeline stops and waits. It never auto-passes.
Strictly opt-in and always last: ownership-level leadership, LinkedIn, email and phone — for approved companies only.
Example theses
"Commercial landscaping companies in Texas, owner-operated, 50+ staff, no private-equity ownership." Searched from the Evergreen Index, PE/franchise-screened, thesis-scored, exported with websites + provenance.
"Founder-owned HVAC and plumbing businesses within 90 miles of Tampa, with company profiles." Multi-wave local discovery + size/HQ/leadership + a fit score against your thesis.
"Independent veterinary clinics in the Southeast, excluding corporate consolidators, ranked by acquisition fit." Discovery + blocklist screening + QA ownership re-verification before delivery.
Install
Evergreen is a single MCP endpoint with 15 tools — no plugin, no skills, nothing to download. Every client gets the full pipeline — sessions, stages, QA and exports live on the server, so there is nothing to install and no files on disk. You’ll need an Acqwired API key (request one).
Add the MCP endpoint as a custom connector. The same steps work on claude.ai in a browser.
Open Settings → Connectors
Connect and authorize
evergreen_…) and click Authorize. The key is exchanged once for a session token and is not stored by Claude.Seeing “Contact an organization owner to install connectors”? Your Claude Desktop account is on a managed org plan. Ask your org admin to enable connector installation in Settings → Organization → Permissions, or use a personal Claude Desktop account instead.
iOS and Android connect to the same MCP endpoint as a custom connector — no plugin, no marketplace. Add the URL once, paste your key in the sign-in sheet, and the full Evergreen pipeline is available in any chat on your phone.
Open Settings → Connectors
Open the authentication page
Register the endpoint, then authorize it — that is the whole setup:
claude mcp add --transport http evergreen https://api.acqwired.com/v1/mcp# then, inside Claude Code: /mcp → evergreen → Authenticate# (a browser opens — paste your evergreen_ key once)No browser available? On CI or a shared machine, register it with a static header instead: --header "Authorization: Bearer evergreen_…".
Add the same URL as a remote (streamable HTTP) MCP server. OAuth is discovered automatically — your browser opens once for the key. Clients that do not support OAuth can send the header Authorization: Bearer evergreen_… instead.
https://api.acqwired.com/v1/mcp