Integration approaches
| Approach | How it works | Best for |
|---|---|---|
| Existing integrations | Connect to 500+ business apps your team already uses and let the agent work through the current integration surface. | Common business tools where the current connection path already exists |
| Agent-level MCP access | Expose a configured agent over MCP to compatible AI clients such as Cursor or Claude Desktop. | Using a Pinksheep agent from a verified MCP client path |
| Webhook triggers | Trigger agent runs from external webhooks so the agent receives the event payload and decides the next step. | Real-time event-driven workflows and external system triggers |
Supported integrations
The current integration surface covers the tools most SMB and mid-market teams already use. The safe claim is broad app coverage, not a promise that every custom API has the same setup path.
Where MCP fits today
MCP is the current verified path for exposing a configured Pinksheep agent to compatible AI clients. In Settings, you get the agent-level endpoint, API key flow, and live connection test for the agents you want to use from Cursor or Claude Desktop.
That means MCP is useful when:
- You want to call a configured business agent from Cursor.
- You want the same agent available in Claude Desktop.
- You need a governed, agent-level endpoint instead of a separate client integration per tool.
For custom or niche API work outside the current integration surface, verify the exact connection model first rather than assuming a public no-code builder exists.
API permissions and scopes
The safe rule is to keep the first integration narrow, review the plan before it runs, and keep consequential actions reviewable while the workflow is still being proven.
Broad promises about permissions, managed OAuth behavior, or custom API handling should only be made when the exact stack and connection model are verified.
Frequently asked questions
Do I need to write API integration code to connect Pinksheep to my tools?
For many common business tools, no. Pinksheep connects to 500+ business apps your team already uses. The exact setup still depends on the tool and the workflow, but the current product surface is designed to remove custom integration work where the connection already exists.
What if my tool is not in the current integration surface?
Then the right path depends on the exact tool and the current MCP surface. The safe approach is to keep the first build narrow, verify the connection model, and avoid assuming there is a public no-code builder for every custom API today.
How does MCP fit into integrations?
MCP is the current verified path for exposing a configured Pinksheep agent to compatible clients like Cursor and Claude Desktop. It is not a blanket claim that every custom API can already be wrapped through a public builder surface.
How should I think about permissions and risky actions?
Review the plan before it runs and keep protected actions reviewable. The safe copy line is simple: let the agent gather context and prepare the work, then keep consequential actions under control.
Can external systems trigger agent runs?
Yes. Webhook-triggered runs are part of the current product truth for event-driven workflows.