If you are new to n8n, the first thing worth knowing is that its integrations are not a fixed marketplace of connectors. They are building blocks. You snap them together on a visual canvas to move data between apps, transform it, and now to run AI agents that make decisions on their own. This guide explains what n8n integrations are, how many exist, which ones matter most, how the new AI nodes work, and how to build automations that hold up in production.
Everything here reflects n8n as it stands in 2026, including the AI features introduced in the 2.0 release.
What are n8n integrations?
n8n integrations are the apps, APIs, databases, and AI models you connect inside an n8n workflow. Each one shows up as a node, which is a single step that either starts a workflow (a trigger) or performs an action (send a message, create a record, call a model). You build automations by dragging nodes onto a canvas and drawing connections between them, so data flows from one step to the next.
The important thing to understand is that n8n integrations come in three layers that stack on top of each other:
Official nodes. These are the pre-built, maintained connectors that ship with n8n, covering popular tools like Slack, Gmail, Google Sheets, Notion, Airtable, HubSpot, Stripe, and GitHub.
Community nodes. Open-source connectors built by the community for long-tail and specialized tools. On n8n Cloud you can now install verified community nodes directly from the canvas rather than only on self-hosted setups.
Anything with an API. The HTTP Request node connects to any service that exposes a REST API, and the Code node lets you write custom logic. Between them, the practical number of integrations is effectively unlimited.
That last layer is the real difference between n8n and connector-locked tools. You are never blocked waiting for someone to build a node.
Key takeaways
n8n integrations are visual nodes that act as triggers or actions inside a workflow, not a closed connector marketplace.
n8n ships 500+ official integrations, backed by 600+ community nodes (nearly 2,000 published on npm), plus unlimited API connections through the HTTP Request node.
The 2.0 release turned n8n into an AI agent platform, adding 70+ dedicated AI nodes with native LangChain support.
Custom logic now lives in the single Code node (JavaScript or Python). The old Function and Function Item nodes are retired.
n8n prices by workflow execution, not per task, which often makes high-volume automation far cheaper than task-based tools.
How many integrations does n8n have?
n8n advertises more than 500 official integrations on its own site, and that number grows almost every month, often driven by user votes and community pull requests. On top of that sit 600+ community nodes, with close to 2,000 packages published on npm that have been downloaded more than 8 million times combined.
You will see larger figures quoted elsewhere, sometimes over 1,000 or even 1,500 "nodes." Those counts usually fold in utility nodes (things like If, Switch, Merge, and Wait) and community packages alongside true app connectors, so treat any single headline number with a little skepticism. The honest way to think about it is this: a few hundred polished official connectors, several hundred more from the community, and an open door to any REST API beyond that.
For most real projects, the count matters less than the flexibility. If your CRM, internal tool, or niche payment gateway is not on the official list, you can still connect it within an hour using the HTTP Request node.
How do n8n integrations work?
Every workflow starts with a trigger. That could be a new form submission, an incoming webhook, a schedule, an updated Airtable row, or a completed Stripe payment. Once the trigger fires, data moves through the nodes you have wired together, and each node reads the output of the one before it.
You shape the flow with logic nodes. If and Switch branch on conditions, Merge joins paths back together, Filter drops items you do not want, and Wait paces requests. Data passes between nodes as JSON, and n8n shows you the output of every step right next to its settings, so you can see exactly what each node received and produced.
A useful habit early on: run a single node on its own before connecting it to anything. n8n lets you execute just that step, inspect the data shape, and confirm the fields are what you expect. This makes debugging far easier than running a whole workflow and guessing where it broke.
The best n8n integrations in 2026
Some nodes show up in almost every workflow. Here are the ones worth learning first, each with a concrete example of how they fit into a real automation.
Slack
Used for notifications and team updates. Example: when a Stripe payment succeeds, post a message to your revenue channel with the customer name and amount, so the team sees new sales in real time.
Google Sheets
A flexible place to log and organize data. Example: append every new lead from a website form to a tracking sheet, then have an AI step write a one-line summary of what the lead asked for.
Airtable
Better than a spreadsheet when you need structured records and relations. Example: create a new record when a customer submits a support form, enrich it with an AI classification (billing, bug, feature request), and route it accordingly.
Notion
Strong for documentation, content, and task tracking. Example: turn each customer interview transcript into a formatted Notion page with key quotes pulled out by a language model.
OpenAI, Anthropic, and other LLM providers
The AI nodes that power text processing, classification, and summarization. Example: read incoming customer messages, categorize sentiment and topic, and pass the structured result to your CRM. n8n connects to OpenAI, Anthropic Claude, Google AI, Hugging Face, and local models through Ollama.
GitHub and GitLab
Popular for automating engineering busywork. Example: open an issue automatically whenever a bug report arrives through a form, with the reporter details and severity already filled in.
Stripe
Central to any SaaS or store. Example: on a successful payment, add the customer to your CRM, log the event in a sheet, and kick off an onboarding email sequence.
HubSpot
The workhorse for marketing and sales flows. Example: create a contact from a form submission, update its lifecycle stage, and attach an AI-generated lead score before a salesperson ever sees it.
HTTP Request and Webhooks
The two most flexible nodes in n8n. Webhooks receive data from any external system, and the HTTP Request node calls any REST API. Example: accept a payload from an internal tool that has no n8n node, transform it, and fan the result out to three other apps.
n8n integration categories
n8n groups its connectors into categories, and it helps to think about them by the job they do inside a workflow rather than by directory label.
Productivity and files. Google Sheets, Notion, Airtable, Microsoft 365, and Google Drive. Used for logging submissions, building structured records, and storing exports and generated documents.
Data and databases. Postgres, MySQL, MongoDB, Redis, and the HTTP Request node. Used for cleaning and validating incoming data, restructuring API responses, and preparing data for analytics or a model.
Communication. Slack, Discord, Telegram, and email. Used for alerts on payments or errors, scheduled reports, and notifications when an AI step finishes.
Developer and project management. GitHub, GitLab, Jira, and Asana. Used for creating issues from reports, syncing pull requests, and pushing metrics into a dashboard.
Marketing and CRM. HubSpot, Mailchimp, Google Analytics, and ad platforms. Used for syncing contacts, triggering campaigns, and enriching leads with AI before storage.
E-commerce and payments. Stripe, Shopify, and WooCommerce. Used for post-payment steps, order and inventory updates, and onboarding sequences.
AI and vector stores. OpenAI, Anthropic, Google AI, Pinecone, Qdrant, and Supabase pgvector. Used for reasoning, retrieval, and memory in agent workflows, covered in detail next.
n8n AI integrations and the AI Agent node
This is the biggest change to n8n in the last year, and it is where the platform now stands apart. The 2.0 release, which landed in January 2026, added more than 70 dedicated AI nodes and made LangChain a first-class part of the canvas. In plain terms, you can build a working AI agent by connecting a few nodes, without writing application code.
The four building blocks of an n8n AI agent
Every agent workflow is built around the AI Agent node, which acts as the brain. You attach sub-nodes to it, and four types matter most:
Model. The language model that does the reasoning. Connect OpenAI, Anthropic Claude, Google AI, Hugging Face, or a local model. You can also set a fallback model that takes over if the primary one fails.
Memory. How the agent remembers a conversation. Options range from a simple window buffer for short chats to Redis or Postgres backed memory for context that survives across sessions.
Tools. What the agent can actually do. Tools can be the HTTP Request tool, the Code tool, a database query, or an entire n8n sub-workflow exposed as a callable function. The agent decides which tool to use based on the request.
Vector store. Long-term knowledge for retrieval-augmented generation. n8n supports Pinecone, Qdrant, Weaviate, Supabase pgvector, and Redis, so an agent can look up relevant documents before answering.
The agent runs a reasoning loop: it reads the input, decides whether it needs a tool, calls it, reads the result, and repeats until it can answer. You can see every step of that loop in the execution log, which makes AI behavior far easier to debug than a black box.
A practical AI workflow example
A common starting point is support ticket triage. A chat trigger or email trigger receives the message. The AI Agent node, connected to a model, a memory node, and a vector store holding your help center articles, reads the ticket, retrieves the relevant policy, and either drafts a resolution or routes the ticket to the right person with a short summary attached. Teams that build this pattern often auto-resolve a meaningful share of tier-one tickets and forward the rest with full context.
One lesson from people who run these in production: the description you give each tool matters more than the tool itself. A vague tool description causes more failures than imperfect code, because the agent cannot tell when to use it. Write tool descriptions the way you would brief a new hire.
Multi-agent and MCP
For complex work, the AI Agent Tool node lets one agent call other agents as tools, so you can split a job into a research agent, a writing agent, and a review agent that a supervisor coordinates. n8n can also act as an MCP server, which means other AI systems can call your n8n workflows as tools, and n8n workflows can connect out to MCP servers. This puts n8n in the middle of larger agent stacks rather than off to the side.
The guiding principle across all of this is to mix deterministic steps with AI steps. Use rule-based nodes where the logic is fixed, use the model where judgment adds value, and add a human approval step wherever a wrong decision would be expensive.
Custom code and the HTTP Request node
When built-in nodes do not cover a task, two nodes give you full control.
The Code node lets you write custom JavaScript or Python inside a workflow, with direct access to the data from previous steps. It replaced the old Function and Function Item nodes some time ago, so if you are following an older tutorial that mentions those, use the Code node instead. It offers two modes, Run Once for All Items for batch operations like sorting or aggregating, and Run Once for Each Item for per-record transformations. There is even an Ask AI tab that drafts the code for you as a starting point.
A couple of practical limits are worth knowing. The Code node cannot make HTTP requests or touch the file system directly, so use the HTTP Request node and the Read/Write Files nodes for those. On n8n Cloud, Python runs in a restricted environment without third-party libraries, while self-hosted instances can enable more. n8n 2.0 also tightened security here: code now runs in sandboxed task runners, and environment variable access inside Code nodes is blocked by default on shared instances.
The HTTP Request node is the universal connector. Point it at any REST API, add your credentials or import a curl command, and n8n handles the request. It includes OAuth2 handling, pagination, and response parsing, which saves the boilerplate you would otherwise write by hand. This node is why "does n8n integrate with X" almost always has the same answer: if X has an API, yes.
How to set up an integration in n8n
Adding an integration follows the same five steps every time.
Add the node. Search for the app in the node panel and drag it onto the canvas.
Connect your credentials. Open the node, choose Add Credentials, and authenticate. n8n stores the credential securely and lets you reuse it across workflows without re-entering it.
Configure the action or trigger. Tell the node what to do, for example send data, fetch records, create an entry, or fire on a new event.
Test the output. Run the node on its own and inspect the fields and data shape before you build further. This catches most problems early.
Wire it into the workflow. Connect the node to the rest of your automation, run a full test, and activate the workflow once it behaves as expected.
How to build a reliable n8n workflow
A few habits keep automations stable as they grow.
Start small and testable. Build the core path first, confirm data moves correctly, then extend. A large workflow assembled all at once is painful to debug.
Store secrets as environment variables. Keep API keys and tokens out of node fields. This improves security and makes it easy to move a workflow between staging and production.
Add clear logic and error branches. Filters, conditions, and error handling control how data flows and prevent wasted API calls. They also make the workflow readable to the next person.
Reuse patterns with sub-workflows. If several automations share steps like cleaning input or calling a model, pull those into a sub-workflow you can call from anywhere. It saves time and reduces mistakes.
Keep workflows modular. Small, focused workflows are easier to maintain than one giant automation that tries to do everything.
Document as you go. Every node has a Notes field. A short note on what each step does becomes essential once a workflow touches multiple tools or model calls.
Watch the execution history. The logs reveal repeated failures, rate limits, and unexpected payloads faster than any other tool in n8n.
Troubleshooting common n8n integration problems
Most issues trace back to authentication, node settings, or unexpected data shapes. Here is how to handle the ones you will hit most.
Authentication failures. Tokens expire and permissions change. Reconnect the account, confirm the token scopes, and make sure keys live in environment variables rather than node fields.
Workflows not triggering. Check that the workflow is switched on, verify the webhook URL, and look at the incoming event logs to confirm n8n is actually receiving data.
Unexpected data formats. APIs return data in different shapes, and a change upstream can break a node. Inspect the output and add a Set or Code node to clean or reshape the data before passing it along.
Rate limits. If a service throttles you, add a Wait node between requests or reduce how often the trigger fires.
Single node errors. Open the execution log and run just the failing node. This tells you whether the problem is n8n, the external service, or the data you passed in.
AI-related failures. With AI nodes, errors usually come from model availability, wrong parameters, or payloads that exceed the context window. Confirm the model name, trim the request size, and add a fallback model so the agent degrades gracefully.
Version or migration issues. After the move to n8n 2.0, some behavior changed, especially around Code nodes and environment access. If an older workflow breaks, check Settings and the Migration Report to find steps affected by the update.
The general method is always the same: isolate the failing node, look at the data going in and coming out, and test one step at a time.
How n8n integrations differ from Zapier and Make
If you are choosing between platforms, the integration model is the real dividing line.
Task-based tools like Zapier ship thousands of fully managed connectors, each tested and updated for you. That is genuinely easier for non-technical users, and if an app changes its API, the vendor handles it. The tradeoff is that you are limited to what they have built, and pricing counts every task.
n8n takes the opposite bet. It offers fewer polished connectors but adds the HTTP Request node, the Code node, community nodes, and self-hosting, so your effective integration count is unlimited and your data can stay on your own infrastructure. It also prices by workflow execution rather than per task, which means a single run that touches ten apps still counts as one execution. For high-volume automation, that difference alone can cut costs dramatically, and the self-hosted Community edition is free.
The honest summary: pick a task-based tool if you want the least setup and have no technical help, and pick n8n if you want control, custom APIs, AI agents, and lower cost at scale. If you are weighing the options, our guide to the best n8n alternatives breaks down where each tool fits.
The bottom line
n8n integrations are less a fixed catalog and more a toolkit. You get 500+ official connectors, hundreds of community nodes, and an open door to any REST API, all wired together on a visual canvas. The 2.0 release pushed the platform well past simple app-to-app automation into AI agent territory, with 70+ AI nodes, native LangChain support, memory, vector stores, and multi-agent orchestration. Learn a handful of core nodes, understand how triggers and data flow work, and reach for the Code and HTTP Request nodes when you need to go off the map. That combination is what lets a beginner build something useful in an afternoon and a team run production automation at scale.
To go further, see our guides on building AI agents, Zapier vs n8n, and browse our directory of workflow automation tools to compare n8n with the rest of the field.