Curated protocols, SDKs, and tools for AI agent communication — email, MCP, webhooks, and inter-agent messaging.
Give your agent a real inbox in 4 lines · Python · TypeScript · MCP · Examples
- MCP Servers
- Webhooks and Event Delivery
- Inter-Agent Protocols
- Real-Time Messaging
- Message Queues
- Notification Routing
- Agent-to-Human Communication Patterns
Tools and protocols for giving AI agents full email capability — send, receive, thread management, webhooks.
- Commune — Purpose-built email API for AI agents. Programmatic inboxes, webhook delivery, thread continuity, structured data extraction, prompt injection detection.
pip install commune-mail/npm install commune-ai/uvx commune-mcp - Nylas — Email and calendar API with IMAP/SMTP abstraction. Supports Gmail, Outlook, Exchange, Yahoo. Not agent-native but widely used in agent stacks for reading existing inboxes.
- Resend — Developer-first transactional email API. React Email templates, inbound email parsing, webhook delivery. Clean REST API.
- Inbox Zero — Open-source AI email management. GitHub: elie222/inbox-zero. Self-hostable.
- SendGrid — Twilio's email API. Transactional + marketing. SMTP relay and HTTP API. Not agent-native.
- Postmark — Transactional email API with inbound processing. Known for fastest delivery.
- AWS SES — Email sending + receiving at scale. SNS integration for inbound webhooks. Used by Commune's infrastructure.
- SMTP — Standard protocol for sending email. Agents use SMTP directly via
smtplib(Python) ornodemailer(Node.js). - IMAP — Protocol for reading/organizing email. Used for agents polling existing inboxes.
- MIME — Email format standard. Attachments, HTML/text multipart, encoding.
- mail-parser — Python email parsing library.
- email-reply-parser — Extract just the reply portion from threaded email (GitHub's version).
- flanker — Mailgun's email address + MIME parsing library (Python).
- postal-mime — Email parser for Cloudflare Workers and browsers.
Model Context Protocol (MCP) enables LLMs to use tools via a standardized interface. Servers expose tools to Claude Desktop, Cursor, Windsurf, and any MCP client.
- Commune MCP — Email for Claude Desktop, Cursor, Windsurf. Create inboxes, send email, read threads.
uvx commune-mcp - Gmail MCP — Official Gmail MCP server by Google. Read, send, search Gmail.
- Outlook MCP — Microsoft Outlook via MCP. Part of official MCP reference servers.
- Firecrawl MCP — Web scraping and crawling via MCP. Fetch any URL as markdown.
- Browserbase MCP — Browser automation via MCP. Navigate, click, extract from websites.
- Tavily MCP — Real-time web search via MCP.
- Exa MCP — Semantic web search via MCP.
- E2B MCP — Secure code execution in sandboxes via MCP.
npx @smithery/cli install e2b --client claude
- Supabase MCP — Official Supabase MCP. Query databases, manage tables, run edge functions.
- GitHub MCP — Official GitHub MCP. Read/write repos, issues, PRs, search code.
- Smithery — Primary MCP server registry. Install any server with
npx @smithery/cli install <id> --client claude - MCP.so — MCP server directory and marketplace.
- Awesome MCP Servers — Community curated list. 10,000+ stars.
Infrastructure for reliable webhook ingestion and delivery to agents.
- Svix — Webhook sending service-as-a-service. Retries, signatures, portal, replay. Used by OpenAI, Clerk, Resend.
- Hookdeck — Webhook ingestion and processing platform. Retry, replay, filter, transform incoming webhooks.
- Inngest — Durable background jobs triggered by events. Retries, fan-out, human-in-the-loop.
- Trigger.dev — Event-driven background tasks for TypeScript. Long-running agent steps without timeouts.
- QStash — HTTP message queue and scheduler for serverless. Exactly-once delivery, cron, deduplication.
- Temporal — Durable workflow execution. Long-running agent pipelines that survive failures. Used by OpenAI.
- Soketi — Open-source, self-hosted Pusher-compatible WebSocket server. Real-time events for agents.
Protocols and frameworks for agents to communicate with other agents.
- A2A (Agent-to-Agent) — Google DeepMind's open protocol for AI agents to communicate. Agent discovery, capability negotiation, task handoff.
- MCP (Model Context Protocol) — Anthropic's standard for LLM-tool integration. Increasingly used for agent-to-tool and agent-to-agent comms.
- ANP (Agent Network Protocol) — Open protocol for decentralized agent communication. JSON-LD, W3C DID-based identity.
- LangGraph — Stateful multi-agent graphs. Agents pass structured state through graph edges. Persistent memory.
- CrewAI — Role-based multi-agent crews. Agents collaborate with defined responsibilities and goals.
- AutoGen / AG2 — Microsoft's conversational multi-agent framework. Agents exchange messages to solve tasks.
- OpenAI Agents SDK — Handoffs, guardrails, sessions. Production-ready multi-agent coordination.
- smolagents — Minimal agent framework with code-first inter-agent communication.
- Redis Pub/Sub — Message broadcasting between agent instances.
- RabbitMQ — Message broker for agent task queuing and distribution.
- NATS — Lightweight messaging for distributed agent systems. Sub-millisecond delivery.
- BullMQ — Redis-based job queue for Node.js. Reliable background task handling for agents.
- Celery — Distributed task queue for Python agents. Widely used in CrewAI/LangChain deployments.
- Slack SDK — Official Python SDK for Slack. Agents read/post to channels, respond to mentions.
- Discord.py — Discord bot framework. Agents in Discord servers.
- Telegram Bot API — Telegram bots for agent interaction. Widely used for personal agents.
- Matrix — Open, decentralized real-time communication protocol. End-to-end encrypted. Growing agent adoption.
- WhatsApp Business API — WhatsApp for business-grade agent messaging.
- Apple Push Notification Service (APNs) — Push for iOS agents.
- Firebase Cloud Messaging (FCM) — Cross-platform push. Used for agent-to-mobile-app notifications.
- OneSignal — Multi-channel push: web, mobile, SMS, email. Widely used for agent notification routing.
- Novu — Open-source notification infrastructure. Multi-channel (email, SMS, push, chat, in-app). GitHub: novuhq/novu — 35k+ stars.
Reliable task and message delivery infrastructure for agent pipelines.
- BullMQ — Redis-based queue for Node.js. Retries, rate limiting, delayed jobs, concurrency. Widely used in AI backends.
- Celery + Redis/RabbitMQ — Python distributed task queue. Standard for LangChain/CrewAI production deployments.
- AWS SQS — Managed message queue. Standard FIFO or best-effort ordering. Used at massive scale.
- Google Cloud Pub/Sub — Real-time message streaming. Global, serverless.
- Apache Kafka — High-throughput event streaming. Agent event sourcing at enterprise scale.
- NATS JetStream — Persistent messaging for distributed agent systems.
Multi-channel alert and notification infrastructure.
- Novu — Open-source notification orchestration. Multi-step, multi-channel. 35k+ GitHub stars.
- Courier — Multi-channel notification routing: email, SMS, push, in-app, Slack. Smart routing based on user preferences.
- Knock — Notification infrastructure with user preferences, templates, batching, digest.
- PagerDuty — On-call alerting and incident management. Used by agents for critical alerts.
- OpsGenie — Alert management and on-call scheduling. Agent integration for incident response.
Design patterns for how AI agents communicate with humans.
- Email-first agents — Agent has a real email inbox; humans can reply by email. Thread context is preserved. Recommended for async, non-urgent tasks. See: Commune email-for-agents examples
- Webhook-reactive agents — Agent is dormant until an event fires (new email, new message, scheduled trigger). No polling. More efficient, lower latency.
- Approval-gate agents — Agent generates output, sends for human review via email/Slack before acting. Common in financial and legal workflows.
- Digest agents — Agent batches updates and sends periodic summaries (daily/weekly). Reduces notification fatigue.
- Escalation chains — Agent handles simple cases autonomously; escalates to human on uncertain/high-stakes decisions. Configurable confidence thresholds.
See CONTRIBUTING.md for guidelines on adding entries.
Tools must:
- Be publicly documented with a URL
- Have a clear agent communication use case
- Be actively maintained (last commit within 12 months)