Model Context Protocol (MCP)

import { Badge } from ‘@astrojs/starlight/components’;

Agento integrates external tool servers using the standard Model Context Protocol (MCP). This allows the agent to interact with databases, issue trackers, documentation search tools, and cloud APIs.

Governed MCP Execution

External tools must never execute without oversight. In Agento, all tools provided by MCP servers are wrapped in the harness governance layer:

1. Plan Gates: Mutating tools require an approved plan in Act Mode before execution.
2. LINTO Evaluation: Any file mutation triggered by an MCP tool is evaluated against .linto rules before execution.
3. Session Auditing: All tool calls and responses are logged to the audit ledger.

---

Configuration (mcp.json)

Configure MCP servers in your repository’s .agento/mcp.json (for project-specific tools) or ~/.config/agento/mcp.json (for global tools).

{
  "$schema": "https://agento.dev/schemas/mcp.json",
  "mcpServers": {
    "sqlite": {
      "command": "uvx",
      "args": ["mcp-server-sqlite", "--db-path", "./data/app.db"],
      "disabled": false
    },
    "github": {
      "url": "https://mcp.github.com/sse",
      "headers": {
        "Authorization": "Bearer ${GITHUB_TOKEN}"
      },
      "disabled": false
    }
  }
}

---

Supported Transports

Agento supports both standard MCP transport protocols:

• Subprocess (stdio): Spawns local CLI tools or server processes.
• Remote (sse / HTTP): Connects to remote MCP servers over Server-Sent Events with header authentication.