TopolearnMCP lets you drive Gumloop from the AI assistant you already use, and centralises app sign-ins.
MCP (Model Context Protocol) is an open standard that lets an AI assistant use tools offered by other systems. An MCP server offers the tools; an MCP client (Claude, Cursor and similar apps) uses them.
Gumloop works in both directions:
- As a client: many connectors are MCP-based, and you can attach your own MCP servers to agents. Enterprise accounts can host or proxy (pass through) MCP servers centrally.
- As a server: Gumloop's own MCP server lets an outside assistant control Gumloop.
Using the server: add Gumloop's MCP server address to your client and sign in. The first time, you authorise access in a browser window; a Gumloop API key from your profile is also used. The server offers tool groups for:
- Agents (create and configure) and sessions (start conversations, follow up).
- Skills and artifacts (files).
- Connected apps: tools from MCP servers you've connected in Gumloop, so Gumloop acts as one central sign-in layer for Gmail, Slack, Linear and others.
- Company Brain search, organisation admin (teams, audit logs, exports) and documentation search.
- Flows and runs: find legacy workbooks and flows, start runs and inspect results.
Care points:
- An assistant connected this way acts with your permissions; review what it does, and keep approvals on risky tools.
- Treat API keys like passwords and revoke unused ones.
- For repeated jobs, a Gumloop agent with triggers is usually better than asking an outside assistant each time.
Remember
- Gumloop is both an MCP client (connectors) and an MCP server
- The server exposes agents, sessions, skills, Brain, connected apps and flows
- Assistants act with your permissions; keep keys secret and approvals on