AI agent tooling statistics (October 2026)
Original data · updated 2026-10-17Numbers computed by Keyflow from the official MCP registry, the n8n template API and public GitHub skill repositories. Free to cite with a link to keyflowhub.com/stats.
Key findings
- 42% of the 3,000 most-viewed n8n templates use an LLM (an AI Agent, OpenAI or chat-model node).
- OpenAI is the default LLM in n8n: the OpenAI Chat Model appears in 687 templates, 3× more than the runner-up Google Gemini Chat Model.
- HTTP Request is the single most-used n8n node, in 45% of top templates.
- 21% of MCP servers in the official registry offer a hosted (remote) endpoint; the rest run locally.
- npm is the most common way to ship an MCP server (48%), followed by pypi (26%).
- 47% of MCP servers need an API key or token to run.
- Only 45 of the 1,289 Claude skills Keyflow indexes come from Anthropic; the rest are community-built across 13 repositories.
Which LLMs n8n workflows use
| Chat model node | Templates | Share |
|---|---|---|
| OpenAI Chat Model | 687 | 23% |
| Google Gemini Chat Model | 261 | 8.7% |
| OpenRouter Chat Model | 68 | 2.3% |
| Anthropic Chat Model | 60 | 2% |
| Ollama Chat Model | 22 | 0.7% |
| Groq Chat Model | 17 | 0.6% |
| Mistral Cloud Chat Model | 14 | 0.5% |
| Azure OpenAI Chat Model | 12 | 0.4% |
| DeepSeek Chat Model | 9 | 0.3% |
| xAI Grok Chat Model | 3 | 0.1% |
| AWS Bedrock Chat Model | 1 | 0% |
Most-used n8n nodes
| Node | Templates | Share |
|---|---|---|
| HTTP Request | 1,358 | 45% |
| Code | 961 | 32% |
| AI Agent | 776 | 26% |
| OpenAI Chat Model | 687 | 23% |
| Google Sheets | 642 | 21% |
| Structured Output Parser | 349 | 12% |
| Gmail | 343 | 11% |
| Simple Memory | 309 | 10% |
| OpenAI | 302 | 10% |
| Telegram | 296 | 9.9% |
| Basic LLM Chain | 290 | 9.7% |
| Google Drive | 283 | 9.4% |
| Google Gemini Chat Model | 261 | 8.7% |
| Slack | 188 | 6.3% |
| Airtable | 145 | 4.8% |
| Default Data Loader | 128 | 4.3% |
| Call n8n Workflow Tool | 117 | 3.9% |
| HTML | 116 | 3.9% |
| Send Email | 104 | 3.5% |
| Embeddings OpenAI | 100 | 3.3% |
| Recursive Character Text Splitter | 97 | 3.2% |
| Information Extractor | 84 | 2.8% |
| Notion | 70 | 2.3% |
| OpenRouter Chat Model | 68 | 2.3% |
| Anthropic Chat Model | 60 | 2% |
Sample: the 3,000 most-viewed templates on n8n.io (19M total views, 800 authors).
MCP servers by package registry
| Registry | Servers | Share |
|---|---|---|
| npm | 1,392 | 48% |
| pypi | 750 | 26% |
| oci | 319 | 11% |
| mcpb | 199 | 6.8% |
| nuget | 32 | 1.1% |
| cargo | 14 | 0.5% |
2,923 servers with a public repository in the official MCP registry. 199 have 1,000+ GitHub stars; 601 are hosted. See the full ranking.
Claude skills by repository
| Repository | Skills | Share |
|---|---|---|
| alirezarezvani/claude-skills | 367 | 28% |
| davepoon/buildwithclaude | 320 | 25% |
| K-Dense-AI/scientific-agent-skills | 175 | 14% |
| Orchestra-Research/AI-Research-SKILLs | 96 | 7.4% |
| browser-act/skills | 91 | 7.1% |
| deanpeters/Product-Manager-Skills | 77 | 6% |
| Jeffallan/claude-skills | 65 | 5% |
| anthropics/claude-plugins-official | 27 | 2.1% |
| ComposioHQ/awesome-claude-skills | 19 | 1.5% |
| anthropics/skills | 18 | 1.4% |
| obra/superpowers | 15 | 1.2% |
| sergebulaev/linkedin-skills | 13 | 1% |
| OthmanAdi/planning-with-files | 6 | 0.5% |
Method
MCP data comes from registry.modelcontextprotocol.io joined with GitHub star counts. n8n data comes from api.n8n.io, limited to the most-viewed templates. Skills are every SKILL.md in the repositories listed above. A template or server is counted once per node or registry it uses. Raw entries are browsable on /n8n, /mcp and /skills.