A WhatsApp MCP server your AI agent can already operate.

MCP-native design — agents are first-class users.

Waaru's primary integration surface is MCP. Every operation an AI agent might need has a tool: send a message, schedule a broadcast, create a flow draft, look up a conversation, classify intent, escalate to a human, read campaign analytics. The same primitives the dashboard uses are the primitives the agent uses. This is the practical meaning of 'MCP-native'. We did not build a dashboard and then expose a REST API. We built the primitives and exposed them both ways — to humans in the dashboard and to agents over MCP.

40+ tools covering the full WhatsApp Business surface.

• Messaging — send_text, send_template, send_media, send_buttons, send_list
• Contacts — upsert, tag, segment, search, fetch_record
• Conversations — list, fetch, fetch_history, summarize, classify_intent
• Flows — list, create_draft, publish, run, version_compare
• Broadcasts — create, schedule, send_now, report
• Templates — list, submit_for_approval, status
• Knowledge — index_url, upload_pdf, search, list_indexed
• Analytics — fetch_campaign_report, fetch_inbox_report, fetch_flow_funnel
• Handoff — escalate, assign, resolve
• Resolutions — list, index, edit, delete (Scale plan)

Auth — per-workspace API keys, scoped, revocable.

Every MCP connection is authenticated with a workspace-scoped API key. Keys are revocable from the dashboard. You can issue multiple keys per workspace with different tool allowlists — a read-only analytics key for a reporting agent, a full key for a customer-service agent. There is no shared credential. There is no impersonation risk.

Schemas and docs auto-generated, agent-readable.

Every tool ships with a JSON-schema for its arguments and a structured docstring the agent reads at discovery. Tool descriptions are reviewed by humans, tested against real agent prompts, and updated when behaviour changes. Your agent should never be surprised by what a tool returns.

What this unlocks for your agent.

A typical WhatsApp platform forces you through a flow builder your agent cannot use. Waaru-over-MCP means your agent can do what it already does — research, decide, plan, act — and WhatsApp becomes another tool. Real examples founding members are shipping in beta:

• A Claude-powered sales agent that qualifies a lead in chat, books a slot, syncs to HubSpot, and sends a follow-up template the next day.
• A GPT-4 abandoned-cart recovery agent that personalises the recovery message using order and browse history from Shopify.
• An internal ops agent that any team member can prompt with 'send a check-in to every customer who hasn't responded in 14 days' — and the agent runs the segmentation, drafts the message, and ships the broadcast.
• A Gemini support agent that watches every escalated conversation, drafts the human reply, and sends only after the human approves.