MCP in Gumloop
Two types of nodes, a growing ecosystem of connectors, and a smarter approach to how agents use tools at scale.
Two Ways to Connect: Native Nodes and MCP Nodes
When building a workflow in Gumloop, you’ll see two kinds of nodes:
Native nodes are pre-built for the most common tools — Gmail, Slack, Google Sheets, Salesforce. They’re fast, polished, and handle edge cases out of the box.
MCP nodes are created on the fly. Describe what you need in plain language, and Gumloop generates a custom node from any connected MCP server. These work for anything — internal databases, niche SaaS tools, proprietary APIs.
| Native Nodes | MCP Nodes | |
|---|---|---|
| Best for | Common tools (Gmail, Slack, Sheets) | Custom or niche integrations |
| Setup | Drag and drop, ready to go | Describe what you need, node is generated |
| Flexibility | Fixed to what’s been built | Unlimited — any API, any action |
Use a native node when one exists. Reach for MCP when you need something custom. Over time, popular MCP integrations graduate into native nodes.
One Connection, Two Superpowers: Workflows + Agents
Here’s what makes Gumloop’s MCP story unique: the same infrastructure powers both workflows and agents. Connect an MCP server once — say, your company’s CRM — and it’s instantly available as a node in your flows and as a tool your agents can call.
Gumloop agents are AI reasoning engines armed with MCP tools and Gumloop workflows. They figure out which tools to use and string them together to solve open-ended tasks. The key: agents can also run your workflows as tools, combining agent flexibility with workflow reliability.
| Workflows | Agents | |
|---|---|---|
| Best for | Repeatable, predictable tasks at scale | Open-ended, complex problems |
| How they use MCP | Each tool becomes a node in the flow | Agent picks the right tools on the fly |
| Together | Agents use workflows as tools — flexibility + reliability in one system | |
”Connect your tools once. Use them in workflows for the repeatable stuff. Use them in agents for the complex stuff. Same MCP connection, both superpowers.”
Why Gumloop’s Approach Is Different
Most AI tools use MCP the same way: a chatbot calls tools one by one in real-time, and every result passes through the AI’s working memory (context window). This works for simple tasks, but it breaks down fast:
- Too many tools — connecting dozens of servers means the AI has to read hundreds of tool descriptions before it can even start working
- Data bloat — fetching a large document and passing it to another tool means the full document flows through the AI’s memory twice, wasting tokens and hitting limits
Gumloop takes a different approach called MCP scripting. Instead of having the AI call tools live, the AI writes a reusable script that calls the tools. That script runs independently — outside the AI’s memory — and becomes a node you can use in any workflow.
| Traditional (Chatbot) | Gumloop (MCP Scripting) | |
|---|---|---|
| How it works | AI calls tools live, one by one | AI writes a script, script runs independently |
| Repeatable? | No — logic disappears after the task | Yes — script becomes a reusable node |
| Large data? | Struggles — everything flows through AI memory | Handles it — data stays in the script, not the AI |
| Best for | Quick, one-off questions | Workflows you run again and again |
The result: workflows that don’t just work once — they work every time, at any scale.