Skip to main content

​
What is MCP?

The Model Context Protocol (MCP) is an open standard for connecting AI models to external tools and data sources. It provides a standardized way for AI applications to discover, authenticate, and invoke tools.

​
Pinksheep and MCP

Pinksheep agents can be exported as MCP-compatible tool definitions. This means agents you build in Pinksheep can be consumed by other AI tools that support MCP, including:
  • Cursor
  • Claude Desktop
  • Windsurf
  • Any MCP-compatible client

​
How it works

When you export an agent as MCP, Pinksheep generates a tool definition that includes:
  • Tool name and description derived from your agent’s name and steps
  • Input schema based on the agent’s parameters
  • Execution endpoint pointing to Pinksheep’s execution infrastructure
  • Authentication handled via your Pinksheep API key
The exported agent runs through the same execution pipeline as a normal Pinksheep agent: pre-flight checks, step execution, approval gates, and result capture.

​
Exporting an agent

1

Navigate to your agent

Go to your Dashboard and select the agent you want to export.
2

Open agent settings

Click the settings or menu icon on the agent detail page.
3

Select MCP Export

Choose “Export as MCP” to generate the tool definition.
4

Copy the configuration

Copy the MCP server configuration and add it to your AI tool’s MCP config file.

​
Configuration example

Add the exported agent to your MCP client config: 
{
  "mcpServers": {
    "pinksheep-crm-enricher": {
      "url": "https://app.pinksheep.ai/mcp/agent/{agent_uuid}",
      "headers": {
        "Authorization": "Bearer {your_api_key}"
      }
    }
  }
}

​
API keys

MCP export requires an API key for authentication. Generate one from Settings > API Keys in the Pinksheep dashboard. API keys inherit the permissions of your account and are scoped to agent execution only.
Keep your API key secure. It provides execution access to your agents and their connected tools.

​
Limitations

  • Agents in “propose” mode still require approval for write actions; MCP calls for these agents will return the proposed action for your review
  • Scheduled agents cannot be triggered via MCP (they run on their own schedule)
  • Credit consumption is the same as normal execution