MCP Resources and Prompts: Beyond Tools

Most MCP tutorials stop at tools, but the protocol has three server primitives, each controlled by a different party:

• Tools - the model decides when to call them.
• Resources - the application (or user) decides what data to attach.
• Prompts - the user explicitly picks them, like a slash command.

Static resources

server.registerResource(
  'db-schema',
  'schema://main',
  { title: 'Database schema', description: 'Tables and columns of the main database', mimeType: 'text/plain' },
  async (uri) => ({
    contents: [{ uri: uri.href, text: await describeSchema() }],
  })
);

Resource templates

Templates expose a whole family of resources through a URI pattern:

import { McpServer, ResourceTemplate } from '@modelcontextprotocol/sdk/server/mcp.js';

server.registerResource(
  'customer',
  new ResourceTemplate('customers://{id}', { list: undefined }),
  { title: 'Customer record', mimeType: 'application/json' },
  async (uri, { id }) => ({
    contents: [{ uri: uri.href, text: JSON.stringify(await db.customers.find(id)) }],
  })
);

Prompts

Prompts package a workflow the user triggers by name, with arguments:

server.registerPrompt(
  'review-migration',
  {
    title: 'Review a SQL migration',
    description: 'Check a migration for locking, data loss and missing indexes',
    argsSchema: { sql: z.string() },
  },
  ({ sql }) => ({
    messages: [
      {
        role: 'user',
        content: { type: 'text', text: `Review this PostgreSQL migration for long locks, data loss and missing indexes:\n\n${sql}` },
      },
    ],
  })
);

Choosing the right primitive

• "The model should decide to do this" -> tool.
• "The user should be able to attach this as context" -> resource.
• "Users run this same task often" -> prompt.

Support for resources and prompts varies between host applications, so check what your target hosts display. Tools are the most widely supported.