MCP Connectors

Orbicus integrates with the Model Context Protocol (MCP), an open standard developed by Anthropic that enables models to securely discover and execute external tools, query databases, and read external documentation.


The Model Context Protocol

Rather than hardcoding proprietary tool adapters, Orbicus uses MCP to mount external tools dynamically at runtime:

  • Local Process Transport (stdio): Launches local command-line tools (written in Node.js, Python, Go, etc.) as child processes communicating over standard input and output.
  • Remote Server Transport (sse): Connects to remote services and cloud APIs using Server-Sent Events over HTTPS.
  • Dynamic Tool Discovery: When an MCP server starts, it advertises its available tools, schemas, and resource templates directly to the agent runtime.

Configuring Connectors

In Settings $\rightarrow$ MCP Connectors, you can add and manage servers using JSON configurations.

Local Stdio Example (PostgreSQL)

To give the agent read-only query access to a local development database:

{
  "name": "postgres",
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://localhost:5432/dev_db"],
  "env": {
    "NODE_ENV": "development"
  }
}

Remote SSE Example (Documentation Gateway)

To connect to a corporate internal documentation server:

{
  "name": "internal-docs",
  "url": "https://mcp.company.internal/sse",
  "headers": {
    "Authorization": "Bearer YOUR_MCP_API_TOKEN"
  }
}

Tool Execution & Permissions

When an agent needs to invoke an external MCP tool (e.g. querying a database or creating a GitHub issue):

  1. Tool Identification: The model selects the tool and populates the required arguments based on the server's registered JSON schema.
  2. Approval Verification: Depending on your autonomy settings, Orbicus prompts you with the exact tool call and arguments before sending them to the MCP server.
  3. Execution & Context Return: The external process executes the query and returns the structured result back to Orbicus as conversational context.