This is a developer-focused guide to interacting with superglue’s capabilities through the Model Context Protocol (MCP). MCP allows for a standardized way to list and call tools exposed by a server.

​
introduction

Superglue utilizes MCP to expose its core functionalities as a set of “tools.” This allows external clients or agents to discover and execute these functionalities in a structured manner. You can think of it as an RPC-like mechanism where Superglue acts as the server, offering services like capability management, data transformation, and more.

​
connecting to the MCP Server

The superglue MCP server is typically available at the /mcp endpoint of your superglue instance.

  • Hosted Endpoint: https://mcp.superglue.ai/
  • Self-Hosted Endpoint: http://<your-superglue-host>:<port>/mcp (e.g., http://localhost:3000/mcp)
String for Cursor / Windsurf / Claude Code (requires mcp-remote)
{
	"mcpServers": {
	  "superglue": {
			"command": "npx",
			"args": [
				"mcp-remote",
				"https://mcp.superglue.ai",
				"--header",
				"Authorization:${AUTH_HEADER}"
			],
			"env": {
				"AUTH_HEADER": "Bearer YOUR_SUPERGLUE_API_KEY"
			}	
	    }
	}
}

The reason this request is structured in an odd way (AUTH_HEADER instead of AUTH_TOKEN) is because Cursor does not allow spaces within the args array, but does so for env vars. This might become obsolete in future versions.

​
authentication

All requests to the MCP server must be authenticated. Superglue’s MCP integration uses a key-based authentication system just like the GQL endpoint.

​
session management

MCP interactions are session-based. A session allows the server to maintain context across multiple requests from the same client.

  • Establishing a Session:
    • To start a new session, the client sends an MCP initialize request to the /mcp endpoint (typically via a POST request).
    • The server responds with a sessionId (e.g., in a header or the response body, though MCP standard usually involves the server generating it and the client then using it). The Superglue implementation generates a UUID for the session.
  • Maintaining a Session:
    • For subsequent requests within the same session, the client must include the mcp-session-id header with the value of the sessionId received during initialization.
    • POST /mcp: Used for most MCP requests like listTools and callTool.
    • GET /mcp & DELETE /mcp: The handleMcpSessionRequest in mcp-server.ts suggests these might be used for session-specific operations, requiring the mcp-session-id header. For example, to check session status or explicitly close a session if implemented.

​
available superglue tools via MCP

The following tools are exposed by Superglue’s MCP server. The input schemas are defined using Zod in mcp-server.ts.

​
1. transformData

  • Description: Execute a data transformation using a defined Superglue transform or an ad-hoc instruction.

  • Input Schema: TransformOperationInputSchema

    • input:
      • endpoint: (Optional) Object defining the transformation (id, instruction, responseSchema, responseMapping, version).
      • id: (Optional) ID of a pre-existing transform. (Either endpoint or id must be provided).
    • data: The JSON data to transform.
    • options: Optional RequestOptionsSchema (cacheMode, timeout, retries, retryDelay, webhookUrl).
    • superglueApiKey: Your Superglue API Key. (Handled server-side by createMcpServer).

  • Example Usage (Conceptual MCP Call):

    // MCP callTool params
{
  "toolName": "transformData",
  "inputs": {
    "input": {
      "id": "my-transform-id"
    },
    "data": { "key": "value" }
  }
}

​
2. listCapabilities

  • Description: List Superglue capabilities (workflows) with pagination.

  • Input Schema: ListWorkflowsInputSchema

    • limit: (Optional) Number of items to return (default: 10).
    • offset: (Optional) Offset for pagination (default: 0).

  • Example Usage (Conceptual MCP Call):

    // MCP callTool params
{
  "toolName": "listCapabilities",
  "inputs": {
    "limit": 5,
    "offset": 0
  }
}

​
3. getCapability

  • Description: Get a specific Superglue capability (workflow) by its ID.

  • Input Schema: GetWorkflowInputSchema

    • id: The ID of the capability.

  • Example Usage (Conceptual MCP Call):

    // MCP callTool params
{
  "toolName": "getCapability",
  "inputs": {
    "id": "capability-id-123"
  }
}

​
4. runCapability

  • Description: Execute a Superglue capability (workflow) by its ID.

  • Input Schema: ExecuteWorkflowInputSchema

    • id: The ID of the capability to execute.
    • payload: (Optional) JSON payload to pass to the capability.
    • credentials: (Optional) JSON credentials for the capability execution.
    • options: (Optional) RequestOptionsSchema.

  • Example Usage (Conceptual MCP Call):

    // MCP callTool params
{
  "toolName": "runCapability",
  "inputs": {
    "id": "capability-id-123",
    "payload": { "inputData": "example" }
  }
}

​
5. buildCapability

  • Description: Build a new Superglue capability (workflow) from an instruction.

  • Input Schema: BuildWorkflowInputSchema

    • instruction: Natural language instruction for the capability.
    • payload: (Optional) Example JSON payload.
    • systems: Array of SystemInputSchema defining the systems the capability can interact with.
      • id: System ID.
      • urlHost: Host URL of the system.
      • urlPath: (Optional) Base path for API calls.
      • documentationUrl: (Optional) URL to API documentation.
      • documentation: (Optional) Inline API documentation.
      • credentials: (Optional) Credentials for the system.
    • responseSchema: (Optional) JSON schema for the expected response.

  • Example Usage (Conceptual MCP Call):

    // MCP callTool params
{
  "toolName": "buildCapability",
  "inputs": {
    "instruction": "Fetch user data from system A and send it to system B.",
    "systems": [
      { "id": "systemA", "urlHost": "https://api.systema.com" },
      { "id": "systemB", "urlHost": "https://api.systemb.com" }
    ]
  }
}

​
6. upsertCapability

  • Description: Create a new capability (workflow) or update an existing one.

  • Input Schema: UpsertWorkflowInputSchema

    • id: The ID for the capability (used for creation or to identify the one to update).
    • input: The capability definition (JSON, conforming to Superglue’s workflow structure).

  • Example Usage (Conceptual MCP Call):

    // MCP callTool params
{
  "toolName": "upsertCapability",
  "inputs": {
    "id": "new-or-existing-capability-id",
    "input": { /* ... full capability definition ... */ }
  }
}

​
7. deleteCapability

  • Description: Delete a Superglue capability (workflow) by its ID.

  • Input Schema: DeleteWorkflowInputSchema

    • id: The ID of the capability to delete.

  • Example Usage (Conceptual MCP Call):

    // MCP callTool params
{
  "toolName": "deleteCapability",
  "inputs": {
    "id": "capability-to-delete-id"
  }
}