# n8n: features, protocols, quickstart

## TL;DR

n8n is a fair-code workflow automation platform from n8n GmbH; agents are one part of it. Workflows can include AI Agent nodes that delegate to other agents, a preview Agent Builder adds standalone agents with sub-agents, and MCP works in both directions. It suits teams automating internal business processes who want agent steps beside ordinary integrations.

## Key facts

| Field | Value |
| --- | --- |
| Type | Platform |
| Languages / SDKs | TypeScript, JavaScript, Python |
| License | unknown |
| Pricing model | [Open core](https://n8n.io/pricing/) |
| Orchestration pattern | Supervisor |
| GitHub stars | 206,348 (as of 2026-09-30) |
| GitHub forks | 60,946 |
| Last push | 2026-09-30 |
| Latest release | n8n@2.41.4 |
| Repository | [n8n-io/n8n](https://github.com/n8n-io/n8n) |
| Website | [n8n.io](https://n8n.io/) |
| Documentation | [docs.n8n.io](https://docs.n8n.io/) |
| Last verified | 2026-09-30 |

## Key features

- The AI Agent node runs as a Tools Agent using LangChain tool calling: it takes a chat model sub-node, at least one tool sub-node, an optional memory sub-node and, when a fixed output format is required, an output parser. ([source](https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.agent/tools-agent))
- The AI Agent Tool sub-node turns an agent into a tool of another agent, so a primary agent can hand tasks to specialist agents; these can be nested several levels deep inside one workflow. ([source](https://docs.n8n.io/integrations/builtin/cluster-nodes/sub-nodes/n8n-nodes-langchain.toolaiagent))
- Agents (Preview) are standalone project artifacts with a model, instructions, tools, skills, sub-agents, episodic memory, a knowledge base, Slack/Telegram/Linear channels and cron schedules, plus draft and published versions. ([source](https://docs.n8n.io/build/build-and-manage-agents))
- The MCP Client Tool node lets n8n agents call tools on external MCP servers, with bearer, header or OAuth2 credentials; registry servers can be added from the node panel. ([source](https://docs.n8n.io/integrations/builtin/cluster-nodes/sub-nodes/n8n-nodes-langchain.toolmcp))
- The MCP Server Trigger node exposes the tools and sub-workflows of one workflow to MCP clients over SSE or streamable HTTP. ([source](https://docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-langchain.mcptrigger))
- An instance-level MCP server lets connected clients search workflows, run the ones enabled for MCP, and create or edit workflows and data tables. ([source](https://docs.n8n.io/connect/connect-to-n8n-mcp-server))
- Tools attached to an AI Agent can be routed through a human review step that pauses the run until someone approves or denies the call in Slack, Telegram, n8n Chat or another channel. ([source](https://docs.n8n.io/build/integrate-ai/ai-examples/human-in-the-loop-for-tools))
- Chat memory for agents comes from sub-nodes: a session-window Simple Memory or external stores such as Redis, Postgres and Zep, plus a Chat Memory Manager node for manual control. ([source](https://docs.n8n.io/build/integrate-ai/understand-ai-components/how-memory-works))

## Architecture and orchestration pattern

Pattern: Supervisor.

An n8n workflow is a graph of nodes started by a trigger and run by the n8n server. AI features use cluster nodes: a root node such as AI Agent, with sub-nodes plugged in for the chat model, tools, memory and output parsing. The agent loop runs inside that one node execution, and the result flows to the next node like any other data.

Multi-agent setups are built in two ways. Inside a workflow, a primary AI Agent receives AI Agent Tool sub-nodes as tools and decides which specialist to call; each specialist has its own description, prompt, model and iteration limit, and they can be nested. Outside workflows, the Agents feature (Preview) stores each agent as a separate artifact; an agent can list other published agents as sub-agents, with a configurable cap on how many run in parallel, and workflows can reach published agents through the Message an Agent node.

State lives in the n8n database (SQLite by default, Postgres supported) as workflow definitions and execution records. Workflow agents keep conversation history only through an attached memory sub-node. Standalone agents keep each conversation as a stored session, have session memory on by default, and can enable episodic memory across sessions (this needs an OpenAI credential); the file knowledge base needs n8n Cloud or, when self-hosted, a Daytona sandbox.

### Human in the loop

In workflows, a human review step can be placed in front of any subset of an AI Agent's tools. When the agent picks a gated tool, the run pauses and an approval request with the tool name and parameters goes to the configured channel (for example Slack, Telegram or the n8n Chat interface); approval runs the tool with the model's input, and denial cancels it and tells the agent. The standalone Agents feature has its own approval setting for sensitive tools: the chat shows Approve and Reject, the agent continues after the decision, and pending approvals are visible in the Sessions tab. Agents are edited as drafts and only the published snapshot runs, and a Preview mode lets a person test a draft first.

### Harnesses it can drive

- Claude Code ([evidence](https://docs.n8n.io/connect/connect-to-n8n-mcp-server/mcp-client-examples))
- Codex ([evidence](https://docs.n8n.io/connect/connect-to-n8n-mcp-server/mcp-client-examples))

## Protocols

| Protocol | Support | Evidence | Note |
| --- | --- | --- | --- |
| MCP | Yes (checked 2026-09-30) | [link](https://docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-langchain.mcptrigger) | Server and client: the MCP Server Trigger node serves workflow tools to MCP clients over SSE or streamable HTTP (stdio is not supported), the MCP Client Tool node consumes external MCP servers, and an instance-level MCP server is also documented. |
| A2A | Unknown (checked 2026-09-30) | — | No mention in the docs llms.txt index or README; GitHub code search for a2a in n8n-io/n8n returned only unrelated strings (CSS, HubSpot helpers, schema docs) and none under packages/@n8n/nodes-langchain; no issue titles mention A2A. |
| AG-UI | Unknown (checked 2026-09-30) | — | No mention in the docs llms.txt index or README; GitHub code search for ag-ui in n8n-io/n8n returned nothing, and n8n is not in the AG-UI README integration list. |

## Best for

- Business-process automations that need an LLM agent step next to ordinary app integrations, with approval gates on risky tools.
- Teams that want to run the platform on their own servers for internal use, using the free Community edition. ([shortlist](https://multiagentguide.top/best/self-hosted-local.md))
- Low-code builders who want a supervisor agent delegating to specialist agents on one canvas.
- Letting Claude Code or Codex create and run workflows through n8n's instance-level MCP server.
- Organizations that need SSO, projects, Git-based version control and log streaming, which paid plans unlock. ([shortlist](https://multiagentguide.top/best/enterprise-governance.md))

## Not for

- Products that let their own customers build workflows on n8n, unless a separate commercial agreement is signed with n8n.
- Code-first teams that want an agent library inside their own application rather than a separate workflow server.
- Self-hosted Enterprise installs or queue-mode deployments that need the standalone Agents feature now.

## Quickstart

```sh
curl -fsSL https://get.n8n.io | sh
```

Install not yet verified by this site.

```bash
# Fresh self-hosted install (needs Docker with the compose v2 plugin)
curl -fsSL https://get.n8n.io | sh
# Editor: http://localhost:5678

# Optional, Agents preview (n8n >= 2.32.3): edit ./n8n/.env so that
# N8N_ENABLED_MODULES includes agents, e.g. N8N_ENABLED_MODULES=instance-ai,agents
docker compose -f ./n8n/compose.yml up -d   # restart with the new setting

# In the editor, enable Settings > Instance-level MCP and mark workflows as
# available in MCP. Then register the server in Claude Code
# (replace the host with your instance's public domain):
claude mcp add --transport http n8n https://n8n.example.com/mcp-server/http
# Inside Claude Code, run /mcp and pick "n8n" to finish OAuth.
```

### Common pitfalls

- The one-line setup needs Docker with the `docker compose` v2 plugin; on Windows run it from WSL.
- The docs say `npm install n8n` / `npx n8n` stop working from n8n 3.0 and new installs should use the Docker-based script; npm installs today need Node.js 20.19 to 24.x.
- Agents (Preview) need n8n 2.32.3 or later with `agents` in `N8N_ENABLED_MODULES`; they are not available on self-hosted Enterprise yet, queue mode is not supported for them, the knowledge base needs a Daytona sandbox and channels need a public `WEBHOOK_URL`.
- Instance-level MCP is off until enabled, and each workflow must be enabled for MCP separately; clients connected before agents shipped must reconnect to get the agent tools.
- The MCP Server Trigger node does not support the stdio transport.
- v1 of the AI Agent node (with the agent type setting) is removed in n8n 3.0.

Official quickstart: https://docs.n8n.io/deploy/host-n8n/install-options/one-line-setup

## Pros

- Multi-agent delegation is available without code: AI Agent Tool nodes plug into a primary agent and can be nested. ([source](https://docs.n8n.io/integrations/builtin/cluster-nodes/sub-nodes/n8n-nodes-langchain.toolaiagent))
- Per-tool human approval can be routed to a different channel from the one the user chats in. ([source](https://docs.n8n.io/build/integrate-ai/ai-examples/human-in-the-loop-for-tools))
- MCP is covered as client, per-workflow server and instance-level server, so n8n can both use and serve tools. ([source](https://docs.n8n.io/connect/connect-to-n8n-mcp-server))
- The self-hosted Community edition is free and, per the docs, has almost the complete feature set. ([source](https://docs.n8n.io/choose-how-to-use-n8n))
- Cloud plans are priced by full workflow executions rather than by steps or users, according to the pricing page. ([source](https://n8n.io/pricing/))

## Cons

- The Sustainable Use License is source-available, not open source: use is limited to internal business, non-commercial or personal purposes, and offering n8n to customers needs a commercial agreement. ([source](https://docs.n8n.io/n8n-community-license/community-license))
- Files with .ee. in their name or directory fall under the separate n8n Enterprise License and need a paid license for production use. ([source](https://github.com/n8n-io/n8n/blob/master/LICENSE.md))
- The standalone Agents feature is in Preview, is not yet available on self-hosted Enterprise, and does not support queue mode. ([source](https://docs.n8n.io/build/build-and-manage-agents))
- SSO, projects, environments, external secrets, log streaming and Git version control require a paid plan or license key. ([source](https://docs.n8n.io/choose-how-to-use-n8n))
- npm-based installs stop working from n8n 3.0, so existing npm deployments need to move to the Docker-based setup. ([source](https://docs.n8n.io/deploy/host-n8n/install-options/one-line-setup))

## Alternatives

- [Dify](https://multiagentguide.top/tools/dify.md)
- [Langflow](https://multiagentguide.top/tools/langflow.md)
- [AutoGPT](https://multiagentguide.top/tools/autogpt.md)
- [AnythingLLM](https://multiagentguide.top/tools/anything-llm.md)

## FAQ

### Does n8n support MCP?

Yes, in both directions. The MCP Client Tool node lets n8n agents call external MCP servers, the MCP Server Trigger node exposes a workflow's tools over SSE or streamable HTTP, and an instance-level MCP server lets clients such as Claude Code or Codex build and run workflows.

### Is n8n open source and free?

It is fair-code under the Sustainable Use License, which is source-available rather than open source. Self-hosting the Community edition is free for internal business, non-commercial or personal use; n8n Cloud and the paid Business and Enterprise editions are listed on the pricing page.

### How does n8n handle multiple agents?

In workflows, a primary AI Agent node calls other agents attached as AI Agent Tool sub-nodes, which can be nested. The Agents feature (Preview) also lets a standalone agent delegate to other published agents as sub-agents.

### Does n8n support A2A or AG-UI?

No official support was found. The docs index, README and repository code search on 30 September 2026 showed no A2A or AG-UI integration, and n8n is not in the AG-UI integration list.

### What languages does n8n use?

n8n itself is written in TypeScript on Node.js. Inside workflows, the Code node runs custom JavaScript or Python.

## Sources

- [n8n GitHub repository](https://github.com/n8n-io/n8n)
- [n8n homepage](https://n8n.io/)
- [n8n documentation](https://docs.n8n.io/)
- [n8n pricing](https://n8n.io/pricing/)
- [AI Agent node](https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.agent)
- [AI Agent node: Tools Agent](https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.agent/tools-agent)
- [AI Agent Tool node](https://docs.n8n.io/integrations/builtin/cluster-nodes/sub-nodes/n8n-nodes-langchain.toolaiagent)
- [Build and manage agents](https://docs.n8n.io/build/build-and-manage-agents)
- [MCP Client Tool node](https://docs.n8n.io/integrations/builtin/cluster-nodes/sub-nodes/n8n-nodes-langchain.toolmcp)
- [MCP Server Trigger node](https://docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-langchain.mcptrigger)
- [Connect to n8n MCP server](https://docs.n8n.io/connect/connect-to-n8n-mcp-server)
- [MCP client connection examples](https://docs.n8n.io/connect/connect-to-n8n-mcp-server/mcp-client-examples)
- [Human-in-the-loop for tools](https://docs.n8n.io/build/integrate-ai/ai-examples/human-in-the-loop-for-tools)
- [How memory works](https://docs.n8n.io/build/integrate-ai/understand-ai-components/how-memory-works)
- [Choose how to use n8n](https://docs.n8n.io/choose-how-to-use-n8n)
- [Community license and FAQ](https://docs.n8n.io/n8n-community-license/community-license)
- [LICENSE.md (Sustainable Use License)](https://github.com/n8n-io/n8n/blob/master/LICENSE.md)
- [LICENSE_EE.md (n8n Enterprise License)](https://github.com/n8n-io/n8n/blob/master/LICENSE_EE.md)
- [One-line setup](https://docs.n8n.io/deploy/host-n8n/install-options/one-line-setup)
- [Set up n8n Assistant and enable agents](https://docs.n8n.io/deploy/host-n8n/configure-n8n/set-up-n8n-assistant)
- [Install with npm](https://docs.n8n.io/deploy/host-n8n/install-options/install-with-npm)
- [Code node](https://docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-base.code)

## Unknown fields

license: GitHub reports NOASSERTION, so the field is 'unknown'. LICENSE.md applies the Sustainable Use License (a fair-code, source-available license written by n8n) to the main branch, except files with .ee. in the name or directory, which are under the n8n Enterprise License in LICENSE_EE.md; third-party components keep their own licenses. protocols.a2a and protocols.agui: searched the docs llms.txt index, the README, GitHub code search in n8n-io/n8n and issue titles; nothing official found, and n8n is not in the AG-UI README integration list. pricing_model is set to open-core as the closest value: the free self-hosted edition is source-available rather than open source, with paid Cloud plans and paid self-hosted Business/Enterprise license keys.

Corrections or removal requests: support@multiagentguide.top

---

Data as of 2026-09-30. Not affiliated with listed projects. HTML version: https://multiagentguide.top/tools/n8n