Sanitized, minimal building blocks to grow into the full system. No secrets — you supply your own
via environment variables or a local .env (git-ignored).
The core AI-intelligence pattern, standalone: read a GHL contact, ask Claude for a tier + summary,
print it. Write-back is included but intentionally disabled (wire it through the confirmation flow in
../docs/04-rebuild-guide.md).
export GHL_TOKEN=... GHL_LOCATION_ID=...
python3 profile_review_example.py <contactId>Uses the Claude CLI if installed (flat-rate billing); otherwise prints a representative example.
A minimal MCP tool server: a few read tools + one write tool that queues for confirmation instead
of firing. This is the safe foundation for letting an AI operate your CRM. Extend it with more tools
using the endpoints in ../docs/03-ghl-knowledge-base.md.
pip install mcp requests
GHL_TOKEN=... GHL_LOCATION_ID=... python3 mcp_server_skeleton.pyThen register it with your AI assistant / MCP client.
Follow ../docs/04-rebuild-guide.md stage by stage. These two files cover Stage 2 (the AI brain) and
Stage 3-4 (tools + safe writes). The knowledge base (../docs/03-ghl-knowledge-base.md) has every
endpoint and gotcha you'll need to add the rest.