Week 37 / 2026.
A short record of changes between issues. Full answers and charts live on the question pages.
- Data window
- Sep 7–13, 2026
- Last updated
- Sep 25, 2026
- Author
- Romuald Członkowski
Commentary
Between 7 and 13 September, n8n-mcp produced 118,918 workflows, 16,988 a day on average, across 2,502,508 tool calls with a 0.25% error rate. That is 21% more workflows a day and 40% more tool calls than the previous issue, and the data doesn't show what moved them. There are 154,416 active installs, 10% of them create workflows, and 506 installs hold 31 workflows or more, 3.2% of the creators.
This is the first issue with a retention figure. Of the 12,039 installs that started in the week of 31 August, 57% were still active a week later, which is this window; the other 43% did not come back. The cohort I use has its first week covering the whole window, so this is not a half-filled reading.
Reads make up 60% of tool calls and writes 14%, while fetching executions alone takes 32% of everything. Each workflow created draws 8.3 updates. The window holds 211,807 edits: 2.5 operations each on average, a median of 889 ms, a verifiable improvement in 62% of them. Changing configuration is the most common intent (83,658), adding functionality is two and a half times rarer (32,325), fixing validation accounts for 9,816. Most of the work goes into tuning what already stands.
In this window's corpus (41,593 workflows, median 7 nodes) Code appears in 73%, HTTP Request in 58%, an LLM model node in 7%, an AI Agent in 5%. Data Table is in 9.5% of workflows and has been climbing since March, when it stood at 3.3%. 1.0% of workflows fail validation, and the smaller the workflow, the more often it fails: 2.31% at two or three nodes, 0.75% at four to ten, 0.13% at 101 to 300. The daily rate swings between 0.4% and 3%, so I don't read a single week as a trend.
Changes versus the previous issue
| figure | value | change |
|---|---|---|
| Workflows created | 118,918 | — |
| Workflows per day | 16,988 | +21% |
| Tool calls | 2,502,508 | +40% |
| Calls ending in error | 0.25% | -5% |
| Failed validation | 1.0% | -47% |
| Active installs | 154,416 | +16% |
| Week-1 retention | 57.1% | — |
| Workflows added to corpus (week) | 41,593 | +6% |
This week's answers
- Which LLM powers n8n AI agents: OpenAI, Anthropic, Gemini or Ollama?
In n8n workflows built last week, OpenAI holds 42% of model nodes, Anthropic 19%, Google Gemini 18% and OpenRouter 11%. In Mar 2026 OpenAI had 48% and Anthropic 12%.
- Which n8n nodes do AI agents use most, and which trigger nodes start their workflows?
The most used node in AI-built n8n workflows is Code, present in 73% of last week's workflows, followed by HTTP Request (58%) and Webhook (50%). The median workflow has 7 nodes.
- Which n8n node is growing fastest?
The fastest-growing n8n node is Data Table: its share of AI-built workflows rose from 4.3% in May 2026 to 8.6% in Aug 2026, a 2.0× increase. Convert to File is second (1.8×). Last week Data Table was the most searched node among AI agents.
- Do AI-built n8n workflows handle errors?
Rarely. 2.1% of AI-built n8n workflows from last week contain an Error Trigger node, and 2.0% of those built in Aug 2026. Since March the share has never exceeded 2.5%.
- Do AI agents build n8n workflows from nodes or from code?
From code. The Code node appears in 73% of AI-built n8n workflows from last week and HTTP Request in 58%. In Aug 2026 the Code share was 73%. Agents would rather write JavaScript and call an API than look for a ready-made integration.
- How large are AI-built n8n workflows, and which fail validation?
55% of workflows created last week have at least 11 nodes and 21% have 31 or more. The larger the workflow, the less often it fails validation: 2.3% for 2–3 nodes versus 0.11% for 31–100.
- Which n8n nodes do AI agents search for?
Last week the most searched n8n node was data table (393 installs), ahead of webhook (234) and http request (125). The ranking counts unique installs, not queries.
- How do AI agents (Claude Code, Cursor) build n8n workflows through MCP: build or maintain?
Maintain. Of 2,502,508 tool calls last week, 60% read workflows, executions and lists, and 14% write. Fetching executions alone is 32% of calls. In the same week 118,918 workflows were created, 16,988 a day.
Most searched nodes
data table (393) · Webhook (234) · http request (125) · schedule trigger (120) · google sheets (113) · ai agent (106) · Telegram (96) · Gmail (90) · Code (73) · form trigger (59)
saveExecutionProgress fills disk space
Enabling saveExecutionProgress causes n8n to save after every node execution. On high-volume workflows this rapidly fills disk space. Check available disk before enabling, and only use for critical workflows that need resume capability.