Concepts

MCP

Use MCP content helpers to expose tool results as text, JSON, resources, or tables.

The MCP entrypoint exports content helpers and shared types for turning tool output into model-readable content.

import { json, resource, table, text } from "@tulip-systems/ai/mcp/server";

Content helpers

text() for short human-readable results
json() for structured output
resource() for external or app resources
table() for row-based data

Resolve a tool for MCP

import { json } from "@tulip-systems/ai/mcp/server";
import { resolveMCPTool } from "@tulip-systems/ai/tools/server";
import { createProspect } from "./tool.server";

export const createProspectMCPTool = resolveMCPTool(createProspect, {
  description: "Create a new prospect in the CRM.",
  content: ({ output }) => json(output, { label: "Created prospect" }),
});

Guidance

Keep the canonical tool reusable.
Keep MCP descriptions clear and standalone.
Return compact content that is useful to a model.
Avoid leaking internal implementation details unless they are the point of the tool.

Chat

Stream agent output on the server and render composable chat UI on the client.

Recipes

Concrete patterns for building assistant experiences with Tulip AI.

On this page

Content helpers Resolve a tool for MCP Guidance