MCP

mcp() connects an Agent to external Model Context Protocol servers. It resolves each configured MCP Server and exposes its tools as model-facing Agent tools.

Installation

Import the Capability factory from -hub/agent/capabilities and add it to defineAgent({ capabilities }). Use the configuration example below as the starting point, then tighten modes, policies, stores, and providers for the Agent boundary.

What it adds

The Capability normalizes MCP tool names with the server name, attaches sanitized MCP metadata, and closes MCP clients after the invocation. It can also contribute instructions that name the configured MCP servers.

Configuration

Pass a server map. Each entry can resolve to an MCP client or MCP client configuration owned by the application.

import { defineAgent } from '@vite-hub/agent'
import { mcp } from '@vite-hub/agent/capabilities'
import { docsMcpServer } from '../mcp/docs'

export default defineAgent({
  driver: { model },
  capabilities: [
    mcp({
      servers: {
        docs: docsMcpServer,
      },
    }),
  ],
})

Runtime behavior

During resolution, mcp() connects to each configured MCP Server and asks for its tool set. ViteHub prefixes normalized tool names with mcp_<server>_ and rejects duplicate normalized names.

The Capability redacts secret-shaped metadata keys before exposing MCP metadata.

Requirements

mcp({ servers }) requires a non-empty server map. MCP client configuration uses the optional @ai-sdk/mcp runtime package when ViteHub creates the client from config.

The external MCP Server owns its own credentials, availability, and tool behavior.

Driver support

Inspect and verify

Inspect the Agent tool list in DevTools. MCP tools should use normalized names such as mcp_docs_search.

Run one invocation with a duplicate normalized tool name during development. The Capability should fail before model execution.

Options

Reference

• Official capabilities
• Custom capabilities
• Source: packages/agent/src/capabilities/mcp.ts