Arara MCP

Arara MCP

Turns Claude Code, Claude Desktop, Cursor, Windsurf or ChatGPT into a WhatsApp operator that knows your customers, your templates, your wallet, and your funnel.

Category
Visit Server

README

Arara MCP

npm TypeScript License Docs

Turn Claude Code, Claude Desktop, Cursor, Windsurf or ChatGPT into a WhatsApp operator that knows your customers, your templates, your wallet, and your funnel. Built by AraraHQ — Brazilian CPaaS for WhatsApp, homologated by Meta.

84 tools · OAuth login · stdio + SSE


Install in 30 seconds

Claude Code

claude mcp add arara --scope user -- npx -y ararahq-mcp

Restart Claude Code. In any conversation:

log into arara

Browser opens, you approve, the token lands in your OS keychain. Done.

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "arara": {
      "command": "npx",
      "args": ["-y", "ararahq-mcp"]
    }
  }
}

Restart Claude Desktop. Type "log into arara" in any chat — browser opens, you approve, you're in.

Cursor

Settings → MCP → Add server:

{
  "arara": {
    "command": "npx",
    "args": ["-y", "ararahq-mcp"]
  }
}

Windsurf

.windsurf/mcp.json at project root:

{
  "servers": {
    "arara": {
      "command": "npx",
      "args": ["-y", "ararahq-mcp"]
    }
  }
}

Hosted SSE — no local install

For ChatGPT (Custom GPT), n8n, or any client that speaks SSE:

URL:    https://mcp.ararahq.com/sse
Header: X-Arara-Key: ara_live_xxx

Generate the key at Dashboard → API Keys.

Headless (CI / n8n self-hosted / server)

Skip OAuth — set the env var and run:

ARARA_API_KEY=ara_live_xxx npx ararahq-mcp --stdio

5 things you can do right now

1. Ask the LLM: "Quantas mensagens entregamos essa semana e quanto gastei?" → Calls get_delivery_metrics + get_wallet_balance. Two-line answer.

2. Ask: "Manda a promoção 'black_friday_30off' pros 200 leads mais quentes do mês passado." → Chains list_contactslist_templatesestimate_campaign_cost → approval → create_campaign with idempotency key.

3. Ask: "O cliente +5511999998888 reclamou. Como respondo?" → Reads conversation history with get_conversation_messages, drafts on-brand reply via brain_suggest_reply, sends on approval.

4. Ask: "Tem dinheiro saindo da minha conta? Recupera o que dá." → Scans AbacatePay leaks via find_revenue_leaks, enriches with WhatsApp history, proposes per-leak offer via negotiate_payment.

5. Ask: "Cria um Smart Link pro meu bio do Instagram que mande mensagem pro +5511999998888 com texto pré-pronto." → Calls create_smart_link. Returns short URL with click tracking + QR code.


Why an MCP, not a raw API client?

A raw API client needs you to remember endpoints, payload shapes, validation rules, idempotency keys, the 24h window, template approval state. The MCP wraps all of that into 84 self-describing tools the LLM picks from — plus an OAuth handshake so you never paste keys into your editor.

You speak Portuguese. The LLM does the rest.


Tool index

84 tools across 16 domains. Full descriptions ship as MCP tool metadata (your client surfaces them automatically). High-level breakdown:

<details> <summary><b>Auth</b> — 3 tools</summary>

login · logout · whoami

OAuth device flow: opens browser, polls until approved, stores token in OS keychain (macOS Keychain / Linux Secret Service / Windows Credential Manager). Never on disk.

</details>

<details> <summary><b>Messaging</b> — 4 tools</summary>

send_message · send_template_to_many · get_message_status · list_messages

send_message accepts EITHER text (free text, requires open 24h window) OR templateName + templateVariables (always allowed). Guardian content checks run before dispatch. send_template_to_many handles up to 1000 recipients with per-recipient variables.

</details>

<details> <summary><b>Templates</b> — 6 tools</summary>

list_templates · create_template · get_template_status · get_template_analytics · delete_template · list_paused_templates

create_template supports body + header (text/media/document) + footer + samples. Meta approval typically takes 1–5 minutes; poll with get_template_status.

</details>

<details> <summary><b>Campaigns</b> — 8 tools</summary>

create_campaign · list_campaigns · get_campaign · estimate_campaign_cost · cancel_campaign · create_ab_test · get_ab_test · force_ab_test_winner

create_campaign accepts optional idempotencyKey to safely retry on network errors. estimate_campaign_cost previews spend before commit.

</details>

<details> <summary><b>Conversations</b> — 6 tools</summary>

list_conversations · get_conversation_messages · reply_in_conversation · check_window_status · bulk_window_check · get_lead_stats

reply_in_conversation only works within an open 24h session window. Use send_message with a template to reopen.

</details>

<details> <summary><b>Contacts</b> — 4 tools</summary>

list_contacts · get_contact · upsert_contacts · get_contact_stats

upsert_contacts accepts up to 1000 per call and returns per-row errors so the LLM can summarize bad rows. Custom attributes as free-form JSON.

</details>

<details> <summary><b>Numbers</b> — 7 tools</summary>

list_numbers · get_number_health · request_new_number · list_number_requests · sync_number_with_meta · update_number · get_warming_plan

Manage WhatsApp numbers, request new ones, force sync of profile (display name / vertical) with Meta, track warmup health.

</details>

<details> <summary><b>Smart Links</b> — 4 tools <i>(Arara DNA)</i></summary>

create_smart_link · list_smart_links · update_smart_link · get_smart_link_stats

Trackable wa.me alternative with per-link QR code, click counting, and first-party attribution. Use in bios, ads, footers.

</details>

<details> <summary><b>Brain (AI)</b> — 9 tools</summary>

brain_interact · brain_suggest_reply · get_brain_config · get_brain_metrics · add_brain_knowledge · list_brain_knowledge · update_brain_knowledge · delete_brain_knowledge · ingest_url_to_brain

brain_interact answers about your business using your templates + knowledge + history. brain_suggest_reply drafts on-brand replies. ingest_url_to_brain crawls a URL into the knowledge base.

</details>

<details> <summary><b>Recovery (AbacatePay)</b> — 10 tools <i>(paid feature)</i></summary>

find_revenue_leaks · negotiate_payment · check_payment_status · get_recovery_endpoint · list_recovery_events · configure_recovery_event · test_recovery_event · set_recovery_endpoint_active · list_recovery_ingests · retry_recovery_ingest

Atomic recovery cycles: scan leaks → propose offer → dispatch payment link via WhatsApp → verify payment. Conversation as contract.

</details>

<details> <summary><b>Wallet & account</b> — 5 tools</summary>

get_wallet_balance · list_wallet_transactions · get_organization_info · get_delivery_metrics · get_pricing

Quick reads of the org state. Compose into your own "weekly review" prompt.

</details>

<details> <summary><b>Guardian (brand safety)</b> — 3 tools</summary>

configure_guardian_policy · list_policy_violations · resolve_policy_violation

Custom regex rules per session that block sensitive content + brand-policy violations on outbound. Built-in rules (CPF, CNPJ, CVV, passwords, tokens) are always on.

</details>

<details> <summary><b>Opt-out & LGPD</b> — 6 tools</summary>

list_opt_outs · register_opt_out · revoke_opt_out · check_opt_out · lgpd_export_contact · lgpd_delete_contact

LGPD compliance built in: opt-out lifecycle + data export/delete on request.

</details>

<details> <summary><b>Phone Lookup</b> — 2 tools <i>(R$ 0,05 each)</i></summary>

lookup_phone · lookup_phones_batch

Validate phones before sending. Returns mobile/landline, carrier, hasWhatsapp.

</details>

<details> <summary><b>API keys</b> — 4 tools</summary>

list_api_keys · create_api_key · rotate_api_key · revoke_api_key

rotate_api_key revokes the old and issues new in one call — for handling compromise without downtime.

</details>

<details> <summary><b>Business profile</b> — 3 tools</summary>

get_business_profile · update_business_profile · sync_business_profile

Manage your business display name, description, vertical, and sync state with Meta.

</details>


Authentication

Two modes, picked automatically in this order:

Mode When How
OAuth (device flow) Default for Claude Code, Desktop, Cursor, Windsurf Run login, approve in browser. Token saved to OS keychain. Auto-refresh.
API key via env CI, server, n8n, ChatGPT (SSE), headless ARARA_API_KEY=ara_live_xxx (stdio) or X-Arara-Key header (SSE)

Internal precedence: explicit apiKey arg → SSE session key → OAuth keychain → ARARA_API_KEY env. If none exist, tools fail with MissingAuth and tell you to run login.


Configuration

# OAuth is the default. These are all optional overrides.
ARARA_API_KEY=ara_live_xxx          # Fallback / headless mode
ARARA_BASE_URL=https://...          # Override endpoint (default: https://api.ararahq.com/api)
ABACATE_API_KEY=xxx                 # For AbacatePay tools
ARARA_MCP_TELEMETRY=off             # Disable usage telemetry
PORT=3333                           # SSE mode port (default 3333)
MCP_TRANSPORT=sse                   # Force SSE instead of stdio

Tokens stored via keytar — macOS Keychain, Linux Secret Service, Windows Credential Manager. Never on disk in plaintext.


Local dev

git clone https://github.com/ararahq/ararahq-mcp.git
cd ararahq-mcp
npm install
npm run build
node build/index.js --stdio          # for Claude Desktop / Code testing
PORT=3333 node build/index.js        # for SSE testing on localhost

To wire your local build into Claude Code:

claude mcp add arara --scope user -- node /absolute/path/to/build/index.js --stdio

Get help


Built in São Paulo · MIT license · © AraraHQ

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured