# Loomal in your agent framework.

Working setups for the major agent frameworks — copy the code, paste your API key, ship. Every guide uses the same Loomal MCP server, just wired in the way each framework expects.

## AutoGen

### Python
AutoGen makes multi-agent conversations easy and credentials hard. Loomal fills the gap: each AssistantAgent gets its own identity through an MCP workbench, so revocation is per-agent and the audit trail attributes correctly.

- Per-agent identity
- MCP workbench tools
- AssistantAgent + UserProxyAgent

### [Read guide →](/content/guides/autogen-agent-email/index.html)

## Claude Agent SDK

### TypeScript / Python  
The Claude Agent SDK supports MCP servers natively. Install the Loomal MCP server and your Claude agent gets an addressable email identity, a credential vault, and TOTP — accessible through structured tools the model already knows how to call.

- Send & receive email
- Encrypted vault
- TOTP / 2FA codes

### [Read guide →](/content/guides/claude-agent-sdk-email/index.html)

## Claude Desktop

### MCP config (JSON)  
Claude Desktop reads MCP servers from a config file. Add Loomal in five lines and Claude can send mail, read its inbox, store credentials, and pass 2FA — all from the desktop chat, no code, no extra app.

- No-code MCP install
- Send / receive email from chat
- Vault for API keys

### [Read guide →](/content/guides/claude-desktop-email/index.html)

## CrewAI

### Python  
CrewAI agents play roles — researcher, writer, sales, support. Each role deserves its own identity. Loomal mints one email + vault per agent, so the sales agent's address differs from the support agent's, and revoking one doesn't touch the others.

- Per-role identity
- Crew-wide audit trail
- Email handoffs between agents

### [Read guide →](/content/guides/crewai-agent-email/index.html)

## Cursor

### MCP config (JSON)  
Cursor's chat sidebar speaks MCP. Add Loomal as an MCP server in mcp.json and the AI in your editor can send email, read replies, fetch credentials from a vault, and pass 2FA — useful for triaging inbox without leaving the IDE, or scripting outbound flows in real-time.

- No-plugin install
- Cursor chat tool calls
- Email send / receive

### [Read guide →](/content/guides/cursor-agent-email/index.html)

## LangChain

### Python / TypeScript  
LangChain agents need tools, and email is the tool every agent eventually wants. Wire Loomal in through the langchain-mcp-adapters package and your agent gets a real address, an encrypted vault, and TOTP — as standard LangChain tools the executor can route to.

- BaseTool integration
- Email send / receive
- Vault & TOTP

### [Read guide →](/content/guides/langchain-agent-email/index.html)

## LangGraph

### Python  
LangGraph is the right framework when your agent's job spans hours or days — send an email, wait for a reply, branch on the response. Loomal provides the email primitives as MCP tools that drop into LangGraph nodes, with the agent's own inbox as the durable signal between steps.

- Long-running graphs
- Wait-for-reply nodes
- Stateful checkpointing

### [Read guide →](/content/guides/langgraph-agent-email/index.html)

## Mastra

### TypeScript  
Mastra is TypeScript's most opinionated agent framework — workflows, memory, and tools as first-class primitives. Loomal plugs in as an MCP server, so your Mastra agent picks up email, vault, and TOTP with one registration.

- MCP tool registration
- Agent + workflow support
- Email send / receive

### [Read guide →](/content/guides/mastra-agent-email/index.html)

## n8n

### No-code (HTTP / JSON)  
n8n is the practical no-code home for AI agents. Loomal slots in either as MCP tools on the AI Agent node, or as plain HTTP Request calls in any workflow — your agent gets a real address, an encrypted vault, and TOTP without leaving the canvas.

- AI Agent node MCP support
- HTTP Request fallback
- Webhook-driven inbox

### [Read guide →](/content/guides/n8n-agent-email/index.html)

## OpenAI Agents SDK

### Python / TypeScript  
The OpenAI Agents SDK supports MCP servers as first-class tool sources. Add Loomal as an MCP server and your agent gets a real email address, a credential vault, and TOTP — exposed as native function calls the model can invoke directly.

- mail.send / mail.reply
- mail_list_messages
- vault.get / vault.store

### [Read guide →](/content/guides/openai-agents-sdk-email/index.html)

## Pydantic AI

### Python  
Pydantic AI's appeal is types end-to-end. Loomal's MCP server fits cleanly: every tool comes with a JSON Schema that Pydantic AI converts into validated arguments. Your agent gets email, vault, and TOTP — with types checked before any call hits the wire.

- MCP Server
- Stdio integration
- Typed tool arguments
- Email send / receive

### [Read guide →](/content/guides/pydantic-ai-email/index.html)

## Vercel AI SDK

### TypeScript  
The Vercel AI SDK ships with first-class MCP support via experimental_createMCPClient. Add Loomal as an MCP server and your streamText or generateText calls get email, vault, and TOTP tools the model can invoke directly — no custom tool wrappers.

- streamText with tools
- MCP via SDK 5.x
- Edge & Node runtimes

### [Read guide →](/content/guides/vercel-ai-sdk-email/index.html)

## Don't see your framework?

The Loomal MCP server works with any client that speaks MCP, and the REST API works with anything that speaks HTTP. The patterns above transfer.

### [Read the Docs](https://docs.loomal.ai/)  
### [Browse Glossary](/content/glossary/index.html)

## We’re live on Product Hunt

Loomal gives AI agents an identity — email, credentials, 2FA — all via MCP. An upvote today would mean the world.

### [Upvote us on Product Hunt](https://www.producthunt.com/products/loomal?utm_source=loomal&utm_medium=launch_modal&utm_campaign=ph_launch&ref=loomal_site)