1ai-reach Agent Control Skill

Use this skill when an AI agent needs to operate, inspect, or supervise the 1ai-reach sales pipeline.

What this skill gives you

• A structured MCP control plane via mcp_server.py
• A reusable Python backend in agent_control.py
• Safe read tools for funnel, leads, events, jobs, and integrations
• Controlled write/run tools for pipeline stages, status changes, and live channel tests

Primary rule

Inspect first, act second.

Recommended sequence for agents:

1. get_system_config
2. inspect_integrations
3. get_funnel_summary
4. preview_autonomous_decision or run_stage(..., dry_run=true)
5. only then use live/destructive tools

MCP server

Public HTTPS endpoint (recommended for remote agents)

https://engage-mcp.aitradepulse.com/mcp

Required headers:

Content-Type: application/json
Accept: application/json, text/event-stream

Quick test:

curl -X POST https://engage-mcp.aitradepulse.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

Claude Desktop / local agent (stdio)

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "1ai-reach": {
      "command": "/home/openclaw/.openclaw/workspace/1ai-reach/.venv/bin/python",
      "args": ["/home/openclaw/.openclaw/workspace/1ai-reach/mcp_server.py", "--transport", "stdio"]
    }
  }
}

Start locally as HTTP MCP server

python3 mcp_server.py --transport http --host 127.0.0.1 --port 8766

Local HTTP MCP path:

http://127.0.0.1:8766/mcp

Available tools

Read-only / orientation

• get_system_config
• get_funnel_summary
• list_leads
• get_lead
• get_recent_events
• inspect_integrations
• list_jobs
• get_job
• load_dataframe_snapshot

Safe previews

• preview_autonomous_decision
• run_stage(..., dry_run=true)

Stateful / destructive operations

• run_stage
• start_background_stage
• stop_job
• set_lead_status
• update_lead_fields
• send_test_email
• send_test_whatsapp

Stage names for run_stage

• strategy
• enricher
• researcher
• generator
• reviewer
• blaster
• reply_tracker
• closer
• followup
• sheets_sync
• orchestrator
• autonomous_loop

Typical agent workflows

1. Check system health

1. get_system_config
2. inspect_integrations
3. get_funnel_summary

2. Preview what the autonomous system would do

1. preview_autonomous_decision

3. Generate proposal for one lead

1. get_lead(lead_id=...)
2. run_stage(stage="generator", lead_id="...", dry_run=true)
3. run_stage(stage="generator", lead_id="...", dry_run=false)

4. Start long-running autonomous supervision

1. start_background_stage(stage="autonomous_loop", args=[])
2. get_job(job_id=...)
3. stop_job(job_id=...) when needed

5. Test live delivery channels

1. inspect_integrations
2. send_test_email(...)
3. send_test_whatsapp(...)

Safety notes

• send_test_email and send_test_whatsapp are live sends
• blaster, closer, followup, and non-dry-run orchestrator affect real leads
• Prefer dry_run=true wherever supported
• Use start_background_stage for long-running operations instead of blocking the agent

Backend layout

• agent_control.py — structured control API used by MCP tools
• mcp_server.py — MCP transport/server entrypoint
• scripts/state_manager.py — authoritative SQLite state
• scripts/senders.py — email + WhatsApp delivery
• scripts/autonomous_loop.py — continuous OODA loop

Current channel behavior

• Email uses Brevo/SMTP/Gmail fallback chain from scripts/senders.py
• WhatsApp uses WAHA with hosted/direct fallback and working-session discovery

When not to use this skill

• If you only need to edit implementation code, use normal repository editing workflows
• If you need browser-based dashboards, there is no dedicated web UI yet