Skip to main content
Glama
YAlegend
by YAlegend

B2B Outreach Framework

Give it a company URL. It finds leads and emails them.

No paid subscriptions required. Works with Claude, ChatGPT, Gemini, or a free local AI. Self-improving — gets better with every campaign.


Start Here

Windows

Double-click START_WINDOWS.bat

Mac

Double-click START_MAC.command

First time: right-click → Open (Mac security step, once only)

Linux

bash start_linux.sh

That's it. The wizard guides you through everything.


Related MCP server: coldforge

What It Does

You give it →  company URL  or  a description
                       ↓
          Reads your website
                       ↓
          AI writes your outreach files
                       ↓
          Quality gate checks everything
                       ↓
          Finds matching leads (free test first)
                       ↓
          Sends personalised cold emails
                       ↓
          Monitors replies — improves itself

What You Need

Thing

Where to get it

Cost

Company website or description

You have this

Free

BetterContact account

bettercontact.rocks

Free (40 emails, no card)

A dedicated Gmail

gmail.com

Free

AI (pick one below)

See options

Free

AI Options (all free)

Option

Cost

How

Ollama + Hermes

Free forever

Runs on your computer — wizard installs it

Claude.ai

Free if subscribed

Paste URL into Claude, save the files

Anthropic API

Free $5 credit

console.anthropic.com — no card needed

OpenAI API

Free $5 credit

platform.openai.com — no card needed

The wizard asks you to pick one and sets it up automatically.


After Setup — Commands

# Find 10 leads (free, no email credits used)
openoutreach find 10

# Find 50 leads with verified emails
openoutreach find 50 emails > my-company/leads.csv

# Find + email automatically
openoutreach run 50

# Self-improving watch mode (checks replies every 24h)
python improver.py --slug my-company --mode watch

# Fully autonomous (no prompts, runs forever)
python improver.py --slug my-company --mode auto

The Self-Improving Loop

After sending, the framework monitors replies and improves itself:

  • Reply rate < 3%? → rewrites your email sequence

  • >30% wrong-person replies? → tightens your target ICP

  • Files not specific enough? → rewrites before anything sends

Every change shows you a diff and asks approval — or runs fully autonomously in --mode auto.


Connecting to Your LLM App

Use the framework from Claude Desktop, Cursor, Windsurf, VS Code, or Zed:

python mcp_server.py        # Claude Desktop, Cursor, Windsurf, VS Code, Zed
python mcp_server.py --http # ChatGPT, Gemini (via browser)

Config files for each app are in mcp_configs/. See MCP.md for setup instructions.


Security

  • Your credentials stay on your computer only

  • .env is blocked from GitHub by .gitignore

  • Lead data and reply logs are never committed

  • Gmail uses an App Password, not your real password

  • Full details in SECURITY.md


Files Generated Per Company

my-company/
├── product.md          ← What you sell (OpenOutreach input)
├── target.md           ← Who to target (ICP definition)
├── positioning.md      ← 1-sentence statement + variants
├── email-sequences.md  ← Cold email sequence per vertical
├── linkedin.md         ← Headlines + thought leadership posts
├── discovery-call.md   ← 30-min call script
├── proposal-template.md← Scoped proposal
└── week-by-week.md     ← 6-week execution plan

License

MIT — use freely, commercially or otherwise. OpenOutreach is GPLv3 — fine for personal and business use.

Contributing

Pull requests welcome. Add your company under examples/ once you've run it. See SECURITY.md for reporting security issues.

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    B
    maintenance
    Enables personalized cold email outreach through research, drafting, template management, and deliverability checks, all locally with no required API keys.
    4
    MIT
  • A
    license
    Not graded
    quality
    C
    maintenance
    Enables B2B prospecting from natural language: detect buying signals, score leads against ICP, enrich decision-makers, and draft personalized outreach messages via Claude.
    2
    MIT