@vibe-agent-toolkit/runtime-claude-agent-sdk

Runtime adapter for deploying VAT agents as Claude Agent SDK MCP tools.

Features

- Converts VAT agents to Claude Agent SDK MCP tools
- Supports Pure Function, LLM Analyzer, and Conversational Assistant archetypes
- Maintains conversation history for multi-turn interactions
- Validates inputs and outputs using Zod schemas
- Single-agent or batch conversion patterns

Installation

``bash
npm install @vibe-agent-toolkit/runtime-claude-agent-sdk


or


bun add @vibe-agent-toolkit/runtime-claude-agent-sdk
`

Supported Archetypes

$3

Stateless, deterministic agents with no external dependencies.

`typescript
import { query } from '@anthropic-ai/claude-agent-sdk';
import { haikuValidatorAgent } from '@vibe-agent-toolkit/vat-example-cat-agents';
import { convertPureFunctionToTool } from '@vibe-agent-toolkit/runtime-claude-agent-sdk';

const { server, metadata } = convertPureFunctionToTool(
haikuValidatorAgent,
HaikuSchema,
HaikuValidationResultSchema
);

// Use with Claude Agent SDK
for await (const message of query({
prompt: "Validate this haiku: 'Cat sits on warm mat / Purring in the sunshine / Dreams of tuna fish'",
options: {
mcpServers: { 'haiku-tools': server },
allowedTools: [metadata.toolName]
}
})) {
if (message.type === 'result') {
console.log(message.result);
}
}
`

$3

Single LLM call agents for classification and extraction.

`typescript
import { query } from '@anthropic-ai/claude-agent-sdk';
import { nameGeneratorAgent } from '@vibe-agent-toolkit/vat-example-cat-agents';
import { convertLLMAnalyzerToTool } from '@vibe-agent-toolkit/runtime-claude-agent-sdk';

const { server, metadata } = convertLLMAnalyzerToTool(
nameGeneratorAgent,
NameGeneratorInputSchema,
NameSuggestionSchema,
{
apiKey: process.env.ANTHROPIC_API_KEY,
model: 'claude-3-5-haiku-20241022',
temperature: 0.9
}
);

// Use with Claude Agent SDK
for await (const message of query({
prompt: "Generate a distinguished cat name for an orange cat",
options: {
mcpServers: { 'name-tools': server },
allowedTools: [metadata.toolName]
}
})) {
if (message.type === 'result') {
console.log(message.result);
}
}
`

$3

Multi-turn conversation agents with session state and history.

`typescript
import { query } from '@anthropic-ai/claude-agent-sdk';
import { breedAdvisorAgent } from '@vibe-agent-toolkit/vat-example-cat-agents';
import { convertConversationalAssistantToTool } from '@vibe-agent-toolkit/runtime-claude-agent-sdk';

const { server, metadata } = convertConversationalAssistantToTool(
breedAdvisorAgent,
BreedAdvisorInputSchema,
BreedAdvisorOutputSchema,
{
apiKey: process.env.ANTHROPIC_API_KEY,
model: 'claude-3-5-haiku-20241022',
temperature: 0.7
}
);

// Multi-turn conversation
const turns = [
"Hi! I'm looking for a cat breed. I live in a small apartment.",
"I love classical music! Chopin and Debussy are my favorites.",
"I don't mind grooming. What breeds would you recommend?"
];

for (const turn of turns) {
for await (const message of query({
prompt: turn,
options: {
mcpServers: { 'breed-advisor': server },
allowedTools: [metadata.toolName]
}
})) {
if (message.type === 'result') {
console.log(message.result);
}
}
}
`

Batch Conversion

Convert multiple agents to a single MCP server:

`typescript
import { convertLLMAnalyzersToTools } from '@vibe-agent-toolkit/runtime-claude-agent-sdk';

const { server, metadata } = convertLLMAnalyzersToTools({
generateName: {
agent: nameGeneratorAgent,
inputSchema: NameGeneratorInputSchema,
outputSchema: NameSuggestionSchema,
},
generateHaiku: {
agent: haikuGeneratorAgent,
inputSchema: HaikuGeneratorInputSchema,
outputSchema: HaikuSchema,
},
}, {
apiKey: process.env.ANTHROPIC_API_KEY,
model: 'claude-3-5-haiku-20241022',
temperature: 0.8,
});

// Use all tools together
for await (const message of query({
prompt: "Generate a cat name and haiku",
options: {
mcpServers: { 'cat-llm-tools': server },
allowedTools: [
'mcp__cat-llm-tools__generateName',
'mcp__cat-llm-tools__generateHaiku'
]
}
})) {
console.log(message);
}
`

API Reference

$3

#### convertPureFunctionToTool(agent, inputSchema, outputSchema, serverName?)

Converts a VAT Pure Function agent to Claude Agent SDK MCP tool.

Parameters:
- agent - The VAT pure function agent
- inputSchema - Zod schema for input validation
- outputSchema - Zod schema for output validation
- serverName - Optional server name (defaults to agent name)

Returns: AgentConversionResult

#### convertLLMAnalyzerToTool(agent, inputSchema, outputSchema, llmConfig, serverName?)

Converts a VAT LLM Analyzer agent to Claude Agent SDK MCP tool.

Parameters:
- agent - The VAT LLM analyzer agent
- inputSchema - Zod schema for input validation
- outputSchema - Zod schema for output validation
- llmConfig - LLM configuration ({ apiKey?, model?, temperature?, maxTokens? })
- serverName - Optional server name (defaults to agent name)

Returns: AgentConversionResult

#### convertConversationalAssistantToTool(agent, inputSchema, outputSchema, llmConfig, serverName?)

Converts a VAT Conversational Assistant agent to Claude Agent SDK MCP tool.

Parameters:
- agent - The VAT conversational assistant agent
- inputSchema - Zod schema for input validation
- outputSchema - Zod schema for output validation
- llmConfig - LLM configuration ({ apiKey?, model?, temperature?, maxTokens? })
- serverName - Optional server name (defaults to agent name)

Returns: AgentConversionResult

$3

#### convertPureFunctionsToTools(configs, serverName?)

Batch converts multiple Pure Function agents.

Parameters:
- configs - Map of tool names to agent configurations
- serverName - Server name (defaults to 'vat-agents')

Returns: BatchConversionResult

#### convertLLMAnalyzersToTools(configs, llmConfig, serverName?)

Batch converts multiple LLM Analyzer agents.

Parameters:
- configs - Map of tool names to agent configurations
- llmConfig - Shared LLM configuration
- serverName - Server name (defaults to 'vat-llm-agents')

Returns: BatchConversionResult

#### convertConversationalAssistantsToTools(configs, llmConfig, serverName?)

Batch converts multiple Conversational Assistant agents.

Parameters:
- configs - Map of tool names to agent configurations
- llmConfig - Shared LLM configuration
- serverName - Server name (defaults to 'vat-conversational-agents')

Returns: BatchConversionResult

Types

$3

`typescript
interface AgentConversionResult {
server: ClaudeAgentMcpServer;
metadata: {
name: string;
description: string;
version: string;
archetype: string;
serverName: string;
toolName: string;
};
inputSchema: z.ZodType;
outputSchema: z.ZodType;
}
`

$3

`typescript
interface BatchConversionResult {
server: ClaudeAgentMcpServer;
metadata: {
serverName: string;
tools: Record name: string;
description: string;
version: string;
archetype: string;
toolName: string;
}>;
};
}
`

$3

`typescript
interface ClaudeAgentLLMConfig {
apiKey?: string; // Defaults to ANTHROPIC_API_KEY env var
model?: string; // Defaults to 'claude-3-5-haiku-20241022'
temperature?: number; // Defaults to 0.7
maxTokens?: number; // Defaults to 4096
}
`

Examples

This adapter implements the RuntimeAdapter interface and can be tested with the runtime-agnostic demo:

`bash


See vat-example-cat-agents/examples/runtime-adapter-demo.ts

The common demo works with ALL runtime adapters via the RuntimeAdapter interface

``

For usage patterns, see the code examples in this README above.

Architecture

This adapter:

1. Wraps VAT agents as Claude Agent SDK MCP tools
2. Validates inputs/outputs using Zod schemas
3. Manages conversation history for conversational agents
4. Provides LLM context for agents that need it
5. Returns structured results in Claude Agent SDK format

License

MIT

Related Packages

- @vibe-agent-toolkit/agent-runtime - Core agent definitions
- @vibe-agent-toolkit/vat-example-cat-agents - Example agents
- @vibe-agent-toolkit/agent-skills - Claude Skills adapter