AI Agents Can Think. Now They Need to Act.

Large language models are remarkable at understanding context, writing copy, and reasoning through problems. But ask Claude to send a WhatsApp message to a client, follow up with leads from last week's campaign, or broadcast a product update to your channel -- and it hits a wall. The model can draft the message, but it cannot deliver it. It can plan the campaign, but it cannot execute it.

This is the gap that skills are designed to close. A skill gives an AI agent a set of capabilities it did not have before -- not just knowledge, but the ability to take real actions in the world. Install the right skill, and your agent goes from "here is what I would do" to "done."

MoltFlow publishes a ClawHub skill package that gives any compatible AI agent full WhatsApp automation capabilities. One install. No coding. Thirty-two MCP tools, seven sub-skill modules, ten Python automation scripts, and access to over ninety REST API endpoints.

Here is how it works and why it matters.

What Is ClawHub?

ClawHub is a skill marketplace for AI agents. Think of it as npm for AI capabilities -- a registry where developers publish packaged sets of tools, knowledge, and configuration that any compatible AI agent can install and use immediately.

The concept is straightforward. Instead of manually configuring API endpoints, writing tool definitions, and wiring up authentication for every new capability, you install a skill. The skill handles all of that for you. It tells the agent what tools are available, how to authenticate, what parameters each tool accepts, and what the expected behavior should be.

Skills are not plugins in the traditional sense. They do not inject code into your agent's runtime. A ClawHub skill is a structured package of documentation, tool definitions, and configuration that the agent reads and acts on. The agent itself decides when and how to use each tool based on your natural language instructions.

For Claude Code users specifically, ClawHub skills integrate directly into your workflow. Install a skill, set your environment variables, and start asking Claude to do things it could not do before.

What Is Inside the MoltFlow WhatsApp Skill Package

The MoltFlow skill package is published under two slugs on ClawHub:

moltflow-whatsapp -- the primary package name
whatsapp-automation-a2a -- the legacy slug, kept for backward compatibility

Both slugs point to the same package. Either one works for installation. Going forward, use moltflow-whatsapp.

Seven Sub-Skill Modules

The package is organized into focused modules, each with its own SKILL.md containing endpoint documentation, curl examples, and parameter specifications:

moltflow (Core) -- Session management, messaging, group operations, label syncing, and webhook configuration. This is the foundation everything else builds on.

moltflow-outreach -- Bulk send campaigns with ban-safe throttling, scheduled messages with cron expressions and timezone support, custom contact groups, channel broadcasting, and scheduled reports delivered to WhatsApp.

moltflow-ai -- AI-powered auto-replies in your writing style, RAG knowledge base with PDF and text document support, Whisper voice transcription, and AI-generated reply suggestions.

moltflow-leads -- Lead detection from WhatsApp groups with keyword monitoring, a CRM pipeline with status tracking (new, contacted, qualified, converted, lost), bulk operations, and CSV export.

moltflow-a2a -- Agent-to-Agent protocol support with E2E encrypted messaging, enabling other AI agents to discover and interact with your MoltFlow instance.

moltflow-reviews -- Automated customer review collection, sentiment analysis scoring, auto-approval for positive reviews, and testimonial export as HTML for your website.

moltflow-admin -- Authentication, API key management with scope enforcement, billing integration, usage tracking, and tenant settings.

32 MCP Tools

Beyond documentation, the package exposes 32 MCP tools across seven categories that the agent can call directly:

Category	Tools	Examples
Sessions	5	List sessions, create session, get QR code
Messages	3	Send message, list chats, read chat history
Groups	7	Monitor groups, create custom groups, add members
Leads	2	List leads with filters, update lead status
Outreach	4	Launch bulk send, schedule messages, track campaigns
Usage	2	Check usage against plan limits
Analytics	3	Campaign metrics, contact engagement, delivery history

These tools connect to MoltFlow's native MCP endpoint at apiv2.waiflow.app/mcp using Streamable HTTP -- no npm packages or Node.js required on your machine.

10 Python Automation Scripts

The package references ten ready-to-use Python scripts for common automation workflows:

quickstart.py -- First-run setup and session pairing
send_message.py -- Direct messaging with delivery confirmation
outreach.py -- Custom groups, bulk send, and scheduled message management
leads.py -- Lead pipeline operations and CSV export
group_monitor.py -- Group monitoring setup and keyword configuration
reviews.py -- Review collector setup and testimonial export
ai_config.py -- AI reply configuration and style training
a2a_client.py -- Agent-to-Agent protocol interactions
admin.py -- API key management and tenant administration
gdpr.py -- GDPR data export and contact erasure

Each script authenticates via the MOLTFLOW_API_KEY environment variable and targets the MoltFlow API at apiv2.waiflow.app by default.

Installation: One Command

Installing the MoltFlow skill into Claude Code takes a single command:

bash

clawhub install moltflow-whatsapp

That is it. The skill is downloaded, registered with your Claude Code instance, and immediately available. No build step, no dependency installation, no configuration files to edit.

For the legacy slug:

bash

clawhub install whatsapp-automation-a2a

Both commands install the identical package.

Configuration: API Key Setup

After installation, you need to connect the skill to your MoltFlow account. This requires one environment variable:

bash

export MOLTFLOW_API_KEY="your-scoped-api-key-here"

To get an API key:

Sign up at molt.waiflow.app (free tier available -- 1 session, 50 messages/month, no credit card)
Go to Dashboard > Sessions > API Keys tab
Click Generate API Key
Select a scope preset: Full Access, Read Only, Messaging, Outreach, Leads, or Custom
Copy the key and set it as an environment variable

Use the narrowest scope preset that covers your workflow. If you only need to send messages and check delivery, use the "Messaging" preset. If you are building lead pipelines, use "Leads." Never use full-scope keys with AI agents in shared environments.

Optionally, you can override the API base URL if you are running a self-hosted instance:

bash

export MOLTFLOW_API_URL="https://your-instance.example.com"

The default points to https://apiv2.waiflow.app.

MCP Server Configuration

For direct MCP integration (recommended for Claude Desktop and Claude.ai users), you can also configure the MCP endpoint. The simplest method uses OAuth -- no API key needed:

json

{
  "mcpServers": {
    "moltflow": {
      "url": "https://apiv2.waiflow.app/mcp"
    }
  }
}

Or for Claude Code specifically:

bash

claude mcp add moltflow --transport http --url https://apiv2.waiflow.app/mcp

OAuth handles authentication automatically. On first tool use, you will be prompted to sign in via your browser. All 32 tools become available immediately.

What You Can Do After Install

Here is where it gets practical. Once the skill is installed and your API key is set, you interact with WhatsApp entirely through natural language. No API documentation to read. No curl commands to memorize. Just tell your agent what you want.

Send Messages in Conversation

You: "Send a WhatsApp message to John at +1-555-0123 saying our meeting is moved to 3pm."

The agent calls moltflow_send_message with the phone number (converted to E.164 format) and your message text. It handles session selection automatically if you have one active session, or asks which session to use if you have multiple.

Mine Leads from Groups

You: "Find all leads from the marketing group that came in this week."

The agent calls moltflow_list_leads with a status filter and date range, returning leads captured from your monitored WhatsApp groups. Each lead includes the source group, capture timestamp, phone number, and any detected buying signals.

You: "Which groups are generating the most leads? Show me the top 5."

The agent combines moltflow_list_monitored_groups and moltflow_list_leads to cross-reference group activity with lead volume.

Run Bulk Campaigns

You: "Set up a drip campaign for new signups -- send a welcome message now and a follow-up in 3 days."

The agent orchestrates multiple tools: moltflow_create_custom_group to build the target list, moltflow_create_bulk_send for the immediate welcome message, and moltflow_create_schedule for the follow-up. All outbound messages go through MoltFlow's anti-spam safeguards -- typing simulation, random delays, and rate limiting -- so your account stays safe.

Check Campaign Analytics

You: "Show me delivery stats for last week's campaign."

The agent calls moltflow_get_campaign_analytics and returns a breakdown: messages sent, delivered, read, replied, failed, and ignored. You get delivery rate, read rate, and reply rate at a glance. For real-time monitoring during active campaigns, moltflow_get_bulk_send_status provides live progress via SSE.

Manage the Lead Pipeline

You: "Move all qualified leads from the real estate group into a custom group called Hot Prospects and schedule a follow-up for Monday morning."

This is where the power of having 32 tools in one skill becomes clear. The agent chains moltflow_list_leads (filtered by status and source group), moltflow_create_custom_group, moltflow_add_custom_group_members, and moltflow_create_schedule into a single workflow. What would take fifteen minutes of dashboard clicking happens in one natural language instruction.

Monitor Groups for Buying Signals

You: "Start monitoring the Property Investors group for anyone mentioning budget, viewing, or interested."

The agent calls moltflow_add_monitored_group with your keyword list. From that point forward, any group message containing those keywords automatically creates a lead in your pipeline. You can check new captures anytime by asking "Any new leads from Property Investors?"

Building Custom Workflows

The MoltFlow skill is designed to compose with other tools and skills in your agent's toolkit. Because each MCP tool is a self-contained operation, the agent can combine WhatsApp capabilities with anything else it has access to.

Some examples that become possible when MoltFlow is one skill among several:

WhatsApp plus calendar integration. "Check my calendar for tomorrow's appointments and send a reminder to each client via WhatsApp." The agent reads your calendar, iterates over appointments, and calls moltflow_send_message for each one.

WhatsApp plus data analysis. "Download the latest sales CSV, find customers who spent over $500 this month, and send each one a thank-you message with a discount code." The agent processes the file, builds a recipient list, and executes targeted outreach.

WhatsApp plus web monitoring. "Watch our product page for price changes and broadcast updates to our WhatsApp channel." Combines web scraping or API monitoring with MoltFlow's channel broadcasting tools.

The key insight is that skills are composable. MoltFlow does not need to build integrations with every possible service. It just needs to provide clean, reliable WhatsApp tools. The agent handles the orchestration.

The ERC-8004 Angle: A Verified On-Chain Agent

MoltFlow is not just a skill package -- it is a verified AI agent registered on the Ethereum blockchain under the ERC-8004 standard.

Agent ID: #25477 Chain: Ethereum mainnet (eip155:1) Registry: 0x8004A169FB4a3325136EB29fA0ceB6D2e539a432

What does this mean in practice? When another AI agent needs WhatsApp capabilities, it can discover MoltFlow through on-chain lookup rather than hardcoded configuration. The A2A discovery endpoint at apiv2.waiflow.app/.well-known/agent.json publishes a machine-readable manifest of all available capabilities.

This matters for trust. Any agent can claim to offer WhatsApp automation. An ERC-8004 registration provides cryptographic proof of identity -- the registration transaction, the owner wallet, and the service metadata are all publicly auditable on Ethereum. No impersonation, no ambiguity.

For the ClawHub skill user, this is mostly background infrastructure. But it means the MoltFlow instance you connect to is the real one, verified at the protocol level. As multi-agent systems become more common, this kind of verifiable identity becomes essential.

Security and Scope Control

A few notes on security, since you are giving an AI agent the ability to send WhatsApp messages on your behalf:

Scoped API keys are mandatory. MoltFlow does not allow wildcard API keys. When you create a key, you must select specific scopes. Use the "Messaging" preset if you only need send/receive. Use "Read Only" if your agent should analyze but not act. The scopes field is required -- there is no default that grants everything.

Anti-spam safeguards cannot be bypassed. Every outbound message passes through typing simulation, random delays, burst rate limiting, and reciprocity checks. These protections are server-side. The skill cannot disable them and neither can the agent.

Content filtering is automatic. Outbound messages are scanned for PII leakage, credential patterns, and prompt injection attempts. Messages that fail these checks are blocked before delivery.

Approval mode is available. If you want human review before any AI-generated message is sent, enable require_approval in tenant settings. All AI-generated messages will be held for manual review.

The skill itself does not install packages, run code automatically, or send messages without explicit user confirmation. It is documentation and tool definitions -- the agent reads it and acts on your instructions.

Getting Started in Five Minutes

Here is the full setup from zero to sending your first AI-driven WhatsApp message:

bash

# 1. Install the skill
clawhub install moltflow-whatsapp

# 2. Set your API key
export MOLTFLOW_API_KEY="mf_your_key_here"

# 3. Start Claude Code and ask it to do something
claude

Then, in Claude Code:

You: "List my WhatsApp sessions and send a test message to my own number."

The agent discovers the MoltFlow tools, authenticates with your API key, lists your sessions, and sends the message. If your session is not connected yet, it will walk you through QR code pairing.

From there, you have the full suite: messaging, group monitoring, lead mining, bulk campaigns, scheduled messages, analytics, review collection, and channel broadcasting. All through natural language. All from one install.

Conclusion

The promise of AI agents has always been that they would handle the tedious, repetitive work so you can focus on decisions that matter. But agents without tools are just sophisticated text generators. They can advise, but they cannot execute.

ClawHub skills close that gap. Install moltflow-whatsapp, set an API key, and your agent gains the ability to manage your entire WhatsApp operation -- from sending individual messages to running multi-step outreach campaigns with lead tracking and delivery analytics.

One skill. Thirty-two tools. Ninety-seven API endpoints. And it starts with a single command.

bash

clawhub install moltflow-whatsapp

Ready to get started? Sign up for a free MoltFlow account and install the ClawHub skill to give your AI agent WhatsApp superpowers today. The free tier includes 1 session and 50 messages per month -- no credit card required.