MCP Tools & Agent Skills

Veriprompt supports three prompt classes that enable different interaction patterns with AI models:

Type	Purpose	Use Case
Prompt	Traditional system + user prompt	Chat, completions, analysis
MCP Tool	Model Context Protocol tool	Agent tool calling, external integrations
Skill	Agent skill for Claude Agent SDK	Reusable agent capabilities with trigger patterns

Creating prompts by type

Navigate to Prompts > Create and select the type using the Store As selector:

• Prompt — standard prompt with system and user template
• MCP Tool — adds tool name, input/output schemas, execution mode
• Skill — adds skill category, trigger pattern, required context, output format

Using examples

When you select MCP Tool or Skill, an expandable Examples panel appears below the type selector. Each example shows:

• For MCP Tools: tool name, execution mode, input/output JSON schemas
• For Skills: skill category, trigger pattern, required context

Click Use this example to pre-fill the form with a working template you can customize.

MCP Tool configuration

MCP Tools expose prompts as callable tools for AI agents via the Model Context Protocol.

Key fields

Field	Description	Example
Tool Name	Unique snake_case identifier	github_pr_review
Execution Mode	How the tool runs	AUTONOMOUS, SUPERVISED, DRY_RUN
Input Schema	JSON Schema for tool inputs	{"type": "object", "properties": {...}}
Output Schema	JSON Schema for tool outputs	{"type": "object", "properties": {...}}
Risk Level	Security classification	LOW, MEDIUM, HIGH, CRITICAL

Execution modes

• Autonomous — tool runs without human approval
• Supervised — requires approval before execution
• Dry Run — returns what would happen without executing

Example: Database Query Tool

{
  "toolName": "database_query",
  "executionMode": "DRY_RUN",
  "inputSchema": {
    "type": "object",
    "properties": {
      "schema": { "type": "string", "description": "Database schema DDL" },
      "question": { "type": "string", "description": "Natural language question" },
      "dialect": { "type": "string", "enum": ["postgresql", "mysql", "sqlite"] }
    },
    "required": ["schema", "question"]
  }
}

Skill configuration

Skills are reusable agent capabilities designed for the Claude Agent SDK. They extend MCP Tools with additional metadata that helps agents decide when and how to invoke them.

Key fields

Field	Description	Example
Skill Category	Classification for the skill	code-review, data-analysis, testing
Trigger Pattern	Natural language description of when to use	"Use when the user asks for code review..."
Required Context	JSON defining what inputs the agent must provide	{"source_code": "string", "language": "string"}
Output Format	Expected output type	Markdown, JSON, Code

Available skill categories

• Code Review, Data Analysis, Testing, Documentation
• Code Generation, Debugging, Research, Content Creation
• Translation, Summarization, Classification, Extraction
• Validation, Transformation, Custom

Example: Code Review Skill

{
  "skillCategory": "code-review",
  "triggerPattern": "Use when the user asks for code review or mentions reviewing code changes",
  "requiredContext": {
    "source_code": "string — the code to review",
    "language": "string — programming language",
    "review_type": "enum[security, performance, style, comprehensive]"
  },
  "outputFormat": "Markdown"
}

Browsing examples

Guru Template Library

Navigate to Guru > Templates and use the category filters:

• MCP Tools — browse public MCP tool templates
• Skills — browse public skill templates

Each template card shows type-specific information (tool name, execution mode, skill category, trigger pattern). Click Use to send to the Guru optimizer, or Copy to copy the prompt content.

Studio Repository

Navigate to Studio > Repository and use the tabs:

• My Prompts — your standard prompts
• Skills — your agent skills
• MCP Tools — your MCP tool definitions
• Template Library — public templates to clone

Architect auto-generation

When using the Project Architect, MCP tools and skills are automatically generated based on your project description. The AI analyzes your requirements and proposes:

• Which prompts should be standard prompts vs. MCP tools vs. skills
• Appropriate input/output schemas for tools
• Skill categories and trigger patterns
• An MCP architecture summary with orchestration notes

You can review and customize all generated items before accepting the proposal.

Seeding example data

To populate your instance with example MCP tools and skills:

npm run seed:mcp-skill-examples

This creates 4 MCP Tool and 4 Skill example templates with real, working content that appear in the Guru template library.