Visual workflow builder, multi-agent Swarm, and a first-party VS Code surface — for developers composing AI agents that do real work in real systems via the Model Context Protocol.
n8n has a basic MCP client node. Langflow and Flowise have none. Most "agent platforms" treat tool calls as glorified HTTP. L2M treats MCP, multi-agent topology, and IDE composition as first-class primitives.
stdio, http_streamable, and mock transports ship in-box. The MCP Tool node probes any server, lists its tools, and dry-runs them inline. Expose your own workflows back out as MCP tools that other agents — IDE, CLI, anything MCP-compatible — call by URL.
MCP integration docs →Compose Supervisor → Worker hierarchies via dedicated attachment ports — not chained prompts. Workers become synthetic tools to the parent, recursively. Each worker gets its own model, its own tool set, its own memory.
Agent loop docs →First-party extension turns any L2M workflow into a coding agent inside the editor — streaming responses, patch previews with workspace-path safety, and command approval. Bring your own backend agent topology.
VS Code extension →Every workflow you build in Studio serializes to a versioned JSON document. Check it into git. Diff it. Code-review it. Import it anywhere. The agent runtime, MCP attachments, branching, retries, memory — all of it round-trips through this format.
schemaVersion: "1.0.0" — breaking-change boundary, never silently shifts."nodes": [
{
"id": "trigger",
"type": "webhook_input",
"config": { "path": "agent", "authMode": "hmac_sha256" }
},
{
"id": "agent",
"type": "agent_orchestrator",
"config": {
"systemPromptTemplate": "{{system_prompt}}",
"toolCallingEnabled": true,
"maxIterations": 6
}
},
// + chat_model, memory, mcp_tool nodes...
],
"edges": [
{ "source": "agent", "sourceHandle": "chat_model", "target": "llm" },
{ "source": "agent", "sourceHandle": "tool", "target": "mcp_fs" }
]
No API keys, no Ollama, no setup beyond a 32-byte master key. The bundled echo provider and in-process mock-mcp adapter let you verify the agent runtime end-to-end before swapping in real models and real MCP servers.
Read the 5-minute tutorial →git clone · pnpm installSECRET_MASTER_KEY_BASE64pnpm dev · open localhost:5173The boring stuff that decides whether you can actually ship.
Topological execution with if, switch, try/catch. Template interpolation across nodes via {{key}} from the merged scope.
Session cookies, API keys, TOTP, SAML, LDAP. Project-scoped roles plus custom roles with fine-grained permissions.
AES-256-GCM with master-key rotation via db:export / db:import. External providers: Vault, AWS, GCP, Azure.
HMAC-SHA256, bearer, Stripe, Discord Ed25519, Telegram. Replay window + idempotency keyed on any header.
Prometheus metrics, OTEL tracing, request-ID propagation through logs, dedicated webhook replicas.
Postgres-backed leader election: scheduler + triggers run on the leader, replicas autoscale webhook ingestion, lease handoff tested.
Ten documented topologies — zero-key MCP agent, supervisor swarm, workflow-as-MCP-tool, RAG with citation, structured-output triage, multi-turn artifacts — each with a runnable sample workflow.
helpdesk-agentSlack/Teams + MCP KB + memorysupport-triageAgent → JSON → switch routingswarmSupervisor + multiple workersworkflow-as-mcpExpose flow as MCP toolvscode-reviewerDiff + filesystem MCPrag-with-citationRetriever + cited answerL2M is a pure OSS project — no closed-source enterprise SKU, no SaaS lock-in. Self-host it, fork it, build community nodes, run it as the MCP-native runtime your stack actually needs.