How Agents Use MCPs
Model Context Protocol (MCP) is an open standard that lets AI agents discover and call external tools. In Pinkfish, agents use MCP servers as their toolbox: each server hosts multiple tools (e.g., “read file”, “search email”, “query database”) that the agent can invoke during a conversation. When you configure an agent, you select which MCP servers it can use. The agent then discovers the available tools and calls them as needed to complete tasks. All MCP servers on Pinkfish are hosted and verified by Pinkfish — we do not use third-party MCP servers.Authenticated Integrations
For MCP servers that connect to external services (Google Drive, Slack, Salesforce, etc.), you must first authenticate in Integrations. After you add a connection, your agents can use the tools provided by that integration. For example, connecting to Google Drive enables tools like list files and write files.Sharing and Permissions
The agent acts as your proxy — it only has access to integrations you have permission to use. When you share a custom agent, integrations do not share with it (similar to workflows). If you share a “Google Drive Agent,” recipients must connect their own Google Drive; they cannot use yours unless you also share your connection. When building custom agents, you can suggest in your instructions that users connect specific services (e.g., Google Drive) to get the most from the agent. In the product, use the “Examples and Tools” menu from the agent prompt box to see tools per server.Full Reference
For the complete list of MCP servers, tool schemas, and API details, see the API reference:- Embedded MCP Servers — Built-in servers (code execution, AI models, data storage, search, etc.)
- Application MCP Servers — Integration servers (Google Workspace, Microsoft 365, Slack, Salesforce, and 100+ more)
- MCP Protocol — JSON-RPC format, authentication, and calling patterns