MCP Prompts and Message Flows

Document MCP prompt templates: typed arguments with completion support, expected message flows with role and content type, worked examples with argument-to-output mappings, and prompt engineering patterns for consistent AI behaviour.

February 26, 2026

Prompts are reusable conversation templates that standardise how AI assistants interact with specific domains. Unlike tools (which perform actions) and resources (which provide data), prompts define the structure and flow of AI conversations. They are user-controlled: the end user selects which prompt to activate, and the MCP client renders it with the provided arguments.

When to Use Prompts

Prompts are valuable whenever you want consistent, repeatable AI behaviour across different contexts.

Domain-Specific Conversations

Standardise how the AI discusses technical topics, customer service enquiries, or code reviews. The prompt template maintains consistent tone, format, and information extraction.

Repeatable Workflows

Define step-by-step interaction patterns that the AI follows. Each expected message represents a turn in the conversation with a specific role and content type.

Quality Assurance

Prompt examples serve as test cases. If the AI generates output that deviates from the expected messages, the prompt template provides a quality baseline for comparison.

Arguments

Prompt arguments are named parameters that the user provides when activating the prompt. They are substituted into the expected messages using {argumentName} placeholders.

Property	Type	Purpose
name	string	Unique argument identifier, used as placeholder in templates
description	string	Explains what the argument controls and what values are acceptable
required	boolean	Whether the prompt requires this argument to function
completionSupported	boolean	Whether the server can provide auto-complete suggestions for this argument

Expected Messages

Expected messages define the conversation structure when the prompt is rendered. Each message specifies a role, content type, and either a template or description.

Template Text

The template text field contains the actual message content with argument placeholders. Placeholders use curly brace syntax: {argumentName}.

Based on the following customer preferences, recommend up to
{maxResults} products from the {productCategory} category.

Customer preferences: {customerPreferences}

For each recommendation, provide:
1. Product name and key features
2. Why this product matches their preferences
3. Price and availability
4. Any relevant alternatives

Worked Examples

Prompt examples tie everything together: they show specific argument values mapped to the expected conversation output. Each example includes argument values and the expected messages the AI should produce.

Argument Values

Concrete values for each argument defined in the prompt. These are substituted into the template text to show what the rendered prompt looks like.

Expected Messages

The conversation output for these specific argument values. Shows the role and full content of each message the AI should generate.

Prompt Engineering Patterns

Effective MCP prompts follow several established patterns.

Single Responsibility

Each prompt should do one thing well. Create separate prompts for "analyse code" and "review security" rather than one "analyse and review" prompt.

Structured Output

Define the expected output format in the user message. If you need numbered lists, JSON, or markdown tables, specify the format explicitly.

Guardrails

Include constraints in the system instructions: maximum response length, topics to avoid, required disclaimers, and when to escalate to a human.

Related Guides

MCP Tools and Annotations

Deep dive into MCP tool definitions: annotation flags, safety matrix, input/output schemas, error scenarios, result content types, related resources, and example patterns. Learn how to document tools that AI assistants can use safely and effectively.

MCP Import and Export

Import existing MCP server manifests into NeoArc Studio and export your documented servers as standard JSON. Understand the file format, directory conventions, and how NeoArc maps MCP primitives to its internal data model.