Model Context Protocol (MCP)

Caide features full support for the Model Context Protocol (MCP), the open standard enabling AI assistants to securely connect to external tools, databases, and third-party APIs.

---

Supported Transports

Caide supports both primary MCP communication transports:

1. stdio: Launch local binaries or scripts (e.g. Node.js, Python, Go) that communicate over standard input/output streams.
2. Server-Sent Events (SSE): Connect to remote MCP servers over HTTP/SSE endpoints.

---

Adding an MCP Server

MCP servers can be configured globally in Settings → MCP Servers or scoped per project in .caide/mcp.json:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_token_here"
      }
    },
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-postgres",
        "postgresql://user:pass@localhost:5432/mydb"
      ]
    },
    "filesystem-extra": {
      "command": "node",
      "args": ["/home/user/custom-mcp-tools/index.js"]
    }
  }
}

---

Tool Discovery & Consent Gating

Once an MCP server connects:

• Caide queries its tool definitions and schemas dynamically.
• The tools become available in the model’s turn context alongside built-in file and shell tools.
• Explicit Consent: By default, calls to external MCP tools require explicit approval unless marked as auto-approved in your security settings.