MCP for n8n
Give your n8n workflows the power to act. n8n is yours to run and own. Add UniversalBench as a hosted MCP server and your AI Agent can run real code, search the live web, read and write databases, commit to GitHub, and call any LLM, with safety built in. Own the automation, plug in the power.
Connect UniversalBench to n8n
Open or add an AI Agent
In your n8n workflow, add an AI Agent node, or open one you already have.
Add an MCP Client Tool
Attach an MCP Client Tool node to the agent. This lets the agent use tools from an external MCP server.
Point it at UniversalBench
Set the MCP server URL to your UniversalBench MCP URL and save.
Run the workflow
Your agent can now run code, query databases, search the web, commit to GitHub, and more, right inside n8n.
Your secrets live in an encrypted secret vault
When a capability needs a credential, a GitHub token, a database URL, an API key, you save it once in your secret vault. You enter the secret name on the left and the secret value on the right, then save. It is never shown in the chat and the matching capability unlocks automatically.
Own your automation, add the power
n8n is open source and self-hostable, so you own your workflows and pay no per-task fees. Per-task automation tools charge for every single step, which gets expensive fast on anything complex. UniversalBench adds the heavy lifting that those tools were never built for, real code, databases, live web search, GitHub commits, and any LLM with cost caps, for a fraction of that price. The first 1,000 calls are free, then you pay only per call. Put together, you get complex automation you fully own, at a fraction of per-task pricing.
See it in action
The agent stays in n8n. UniversalBench does the compute-heavy step and hands the result straight back, so the rest of your workflow runs as normal.
Why one server beats ten
Single-purpose MCP servers each do one thing. A GitHub server reads your repo. A database server runs a query. But real work is never one step. You search the web, run code on what you found, write the result to a database, then commit it. No single-purpose server can do that chain on its own. UniversalBench carries the whole combination, so one connection does the job of ten servers wired together, and you keep one URL, one secret vault, and one bill instead of ten of each.
Real work is never one step
Nobody runs just one tool. The moment a task gets real, it needs several working together, and that is exactly where a single-purpose server falls short.
Triage a failing build
Search the web for the error, read the repo, run the failing test, write the fix, and commit it. That one task spans web search, GitHub, and code. A GitHub-only server can read the repo but cannot search the web or run the test. UniversalBench does the whole chain.
Answer a question about your numbers
Query the database, analyze the rows in code, look up a benchmark on the web, and return the figure. A database-only server hands back rows and stops. UniversalBench queries, computes, and looks up the context in one go.
Keep a record current
Find the latest figure online, compute the change in code, write it to the database, and post a note to the team. No single-purpose server spans web, code, database, and messaging. UniversalBench does it end to end.
What it can do once connected
GitHub
read repositories and commit validated code changes
Slack
post messages and read through the Slack API
Postgres
query and update any PostgreSQL-compatible database
Notion
read and update Notion pages and databases
Filesystem
read, write, convert, and analyze files in an isolated sandbox
Web Search
search the live web and return processed results
Google Sheets
read and update Google Sheets
Gmail
read and send Gmail
What UniversalBench adds to n8n
| n8n alone | n8n + UniversalBench | |
|---|---|---|
| Run real code | Basic Code node | Full sandboxed code, validated |
| Live web search | Build it yourself | Built in |
| Databases | Per app nodes | Any PostgreSQL-compatible database |
| Any LLM | Per provider node | Any model, with cost caps on by default |
| Safety | None specific | Code validated, wallet capped, internal network blocked |
Frequently asked questions
How do I connect an MCP server to n8n?
Add an AI Agent node, attach an MCP Client Tool node, and set its MCP server URL to your UniversalBench URL. Save and run the workflow.
Is it cheaper than per-task automation?
It can be. You self-host n8n with no per-task fees and add UniversalBench for the heavy steps, where the first 1,000 calls per month are free and you pay only per call after that.
Does it work with other AIs too?
Yes. The same hosted MCP server works with Claude, ChatGPT, Gemini, Cursor, and any MCP-compatible tool, not just n8n.
