LeadAce

Outbound sales agent for Claude Code: per-prospect research, email writing, Gmail sending, reply tracking, and structured rejection feedback.

Documentation

LeadAce

Status: Public Beta License Listed on mcpservers.org

Autonomous outbound sales agent in your browser. Builds prospect lists, writes and sends one email per company from your own mailbox, and improves its strategy from every reply and rejection.

Website: https://leadace.ai

Two ways to run it. Use the hosted service at app.leadace.ai (Free tier — 30 prospects, paid plans from $49/mo), or self-host the backend on your own Cloudflare + Supabase. The web app and the optional Claude plugin work the same on either.

For Users

Get started

  1. Sign in with Google at app.leadace.ai (Free tier — no card). LeadAce asks to send from your Gmail and to read your inbox, so it can catch the replies.
  2. Paste your website in the chat. Ace proposes who to target and how to approach them. Change anything, then approve.
  3. Read the first drafts. Nothing is sent until you approve, and what you approve goes out from your own inbox.

Email runs entirely in the web app: research, writing, sending and reply collection run on the server. Schedule the daily cycle from the chat or project settings. Further Google accounts and SMTP mailboxes are added in Account settings.

Claude plugin (optional)

The plugin, for Claude Cowork or Claude Code, runs the same work from a Claude session and adds a browser: it sends the contact-form messages and SNS DMs that the web app leaves for you to send by hand. It works on the same projects as the web app.

Prerequisites

  • Claude Cowork (in the Claude Desktop app) or Claude Code, on an Anthropic Pro or Max plan — verified on macOS
  • A LeadAce account at https://app.leadace.ai (Free tier — no card)
  • A connected Gmail account — for sending email (granted when you sign in with Google, or via the "Connect Gmail" banner in the web app)
  • Gmail MCP (claude.ai built-in) — for checking email replies
  • A browser, only for browser channels — contact forms run on Cowork's built-in browser or any browser-automation MCP you configure (e.g. Playwright); SNS DMs and SNS reply checking require Claude in Chrome. A scheduled run reaches the browser only while Claude Desktop is open, with its default browser chosen beforehand

Installation

In Claude Desktop: Customize → Plugins → Add from a repository → aitit-inc/leadace → Install → Connectors → Connect, then Customize → Connectors → LeadAce → Connect and sign in with Google.

Or one line in your terminal (Claude Code):

claude plugin marketplace add aitit-inc/leadace && claude plugin install leadace@leadace

Or, from inside a running Claude Code session:

/plugin marketplace add aitit-inc/leadace
/plugin install leadace@leadace

To update later:

/plugin marketplace update
/plugin update leadace@leadace

Sign in to LeadAce

The first time the plugin calls a LeadAce tool, your browser opens for Google sign-in (the same Google account as the web app). The token is cached locally for subsequent runs. See plugin/README.md for details and troubleshooting.

Usage

Most commands take your project name as the first argument (chosen during /leadace onboarding); /leadace itself takes a free-form question or homepage URL. Run them in a Cowork session (the Cowork tab, not Chat — /daily-cycle runs sub-agents, which Chat can't start) or in Claude Code.

CommandPurpose
Setup
/leadaceEntry point — onboarding, environment setup / re-check, strategy authoring, overview, and routing
Add prospects (pick one)
/build-list <name>Web search for new prospects
/import-prospects <name>Load CSV / Excel / SQLite
/match-prospects <name>Reuse prospects already in your tenant
Sales loop
/outbound <name>Send via email, contact forms, SNS DMs
/check-responses <name>Collect Gmail + SNS replies → DB
/evaluate <name>PDCA — analyse, auto-improve strategy, and surface tactical rejection signals (recontact queue, decision-maker referrals, targeting hints)
Reflection
/check-feedback <name>Surface PMF signals from rejection feedback (feature gaps, competitor presence) — ad-hoc product reflection
Automation
/daily-cycle <name> [count]One-shot bundle: check-responses → evaluate → outbound + build-list
/setup-cron <name>Set up a daily run (a server-side schedule that needs no machine, a Cowork scheduled task, a Claude Code Desktop task, or an OS scheduler)
Maintenance
/delete-project <name>Permanently delete a project and all its data

Projects, prospects, outreach logs, and strategy documents live in the cloud — there are no local files to manage. Review everything in the web app at https://app.leadace.ai.

Flow

flowchart TD
  LA["/leadace<br/>onboard · setup · strategy"] --> P{add prospects}

  P -- web search --> BL["/build-list"]
  P -- CSV / Excel --> IP["/import-prospects"]
  P -- reuse tenant --> MP["/match-prospects"]

  BL --> OB["/outbound"]
  IP --> OB
  MP --> OB

  OB --> CR["/check-responses"]
  CR --> EV["/evaluate"]
  EV -- next round --> P

  CR -. PMF signals .-> CF["/check-feedback"]
  CF -. revisit strategy .-> LA

  DC["/daily-cycle<br/>check + outbound + build, one shot"]
  SC["/setup-cron<br/>daily schedule"] --> DC
  DC -. replaces manual loop .-> P

  DEL["/delete-project"]

Solid arrows = the main loop. Dashed = optional / occasional / wrapper. /evaluate also consumes the tactical slice of rejection feedback (recontact requests, decision-maker referrals, not_relevant industry clusters) recorded by /check-responses — no separate user step.


License

LeadAce is released under the LeadAce Open Source License — a modified Apache 2.0 with two additional conditions:

  • No multi-tenant SaaS for third parties without a commercial license from SurpassOne Inc. Self-hosting for your own organization is fine.
  • Frontend logo and copyright must be preserved in any deployment exposing the LeadAce console.

Hosted service (cloud)

  • Free tier: 30 prospects (lifetime), 1 project, 1 mailbox, 500 stored prospects.
  • Paid plans start at $49/month for 100 prospects a month. A prospect counts once, when the first message to it goes out; follow-ups are free. Manage your subscription from the web app.

Self-host

See docs/self-host.md. The self-hosted edition runs on the unlimited tier — no Stripe, no caps. For commercial-license inquiries, contact leo.uno@surpassone.com.


For Developers

Repository layout

plugin/                          # Claude Code plugin
├── .claude-plugin/plugin.json   # Manifest
├── .mcp.json                    # MCP server config
├── skills/                      # Slash commands (each directory has SKILL.md)
├── scripts/fetch_url.py         # Local web fetch helper
└── references/                  # Shared reference docs
backend/                         # API + MCP servers (Cloudflare Workers, Hono, Drizzle)
frontend/                        # Web app (SvelteKit, Cloudflare Worker + static assets)
docs/                            # Project-wide docs (deploy runbook, self-host, architecture)
docker-compose.yml               # Bare Postgres for non-Supabase local dev

Quick start (local dev)

One-time setup — copy the env templates:

cp backend/.dev.vars.example backend/.dev.vars
cp frontend/.env.example frontend/.env

Fill in the Supabase keys from supabase status — it prints them once the local stack is running, so run make dev once first (it starts Supabase), then paste the keys in.

For Google sign-in to your local stack, also create a Google OAuth client and export SUPABASE_AUTH_EXTERNAL_GOOGLE_CLIENT_ID / _SECRET in your shell (via .envrc) before the first make dev — it boots Supabase, which reads them from the shell at start time. See docs/self-host.md → Local development. (These shell vars gate sign-in; the GOOGLE_CLIENT_ID / _SECRET in backend/.dev.vars are separate — they power Gmail send.)

Then start the whole stack with one command — Supabase, migrations, the master seed, the API/MCP Workers, and the frontend, all together. Ctrl-C tears the dev servers down (Supabase stays up for a fast restart; make stop halts it):

make dev          # or: ./scripts/dev.sh
ServiceURL
Frontendhttp://localhost:5273
API Workerhttp://localhost:8787
MCP Workerhttp://localhost:8788
Supabase Studiohttp://localhost:54323

To run on different ports (e.g. one is taken by another dev server), copy dev.ports.env.example to dev.ports.env and set the ports there — dev.sh rewires every dependent URL (and Google sign-in keeps working). Defaults are unchanged when the file is absent.

Run the steps manually instead
npx supabase start                      # Auth + Postgres on ports 54321/54322
cd backend
npm install
npm run db:migrate
npx tsx scripts/seed-master-documents.ts

npm run dev:api                         # API → http://localhost:8787
npm run dev:mcp                         # MCP → http://localhost:8788  (separate terminal)

cd ../frontend
npm install
npm run dev                             # → http://localhost:5273

Pre-release checks:

cd backend && npm run typecheck
cd frontend && npm run check

Updating dependencies (lockfile gotcha)

npm install with node_modules already present can prune other-platform optional deps (@emnapi/*, @img/sharp-*, esbuild binaries) from package-lock.json (npm/cli#7961, npm 10.3+–11.x). CI then runs npm ci against that pruned lockfile and fails with Missing: … from lock file. This bites both backend/ and frontend/, and is what makes Dependabot's npm PRs go red.

When you change a package.json / package-lock.json (or repair a Dependabot PR), regenerate the lockfile under the repo's pinned toolchain — not in Docker:

nvm use                 # node 24 (repo .nvmrc) — matches CI
cd backend              # or cd frontend
rm -rf node_modules     # removing this first is what avoids the prune
npm install --no-audit --no-fund

Then commit the regenerated package-lock.json. Code-only changes don't need this — CI consumes the committed lockfile as-is.