import Anthropic from "@anthropic-ai/sdk";
const client = new Anthropic();
interface Tool {
name: string;
description: string;
input_schema: {
type: string;
properties: Record<string, unknown>;
required: string[];
};
}
// Minimal in-memory tool definitions
const tools: Tool[] = [
{
name: "query_postgres",
description: "Execute read-only SQL queries on Jarvis Supabase database",
input_schema: {
type: "object",
properties: {
query: {
type: "string",
description: "SQL SELECT query",
},
limit: {
type: "number",
description: "Result limit (default 100)",
},
},
required: ["query"],
},
},
{
name: "search_vectors",
description:
"Semantic search via pgvector on Jarvis knowledge base or memory",
input_schema: {
type: "object",
properties: {
table: {
type: "string",
description: "Table name (e.g., 'documents', 'interactions')",
},
query_text: {
type: "string",
description: "Text to search for (will be embedded)",
},
top_k: {
type: "number",
description: "Number of results to return",
},
},
required: ["table", "query_text"],
},
},
];
// Simulated tool execution
async function executeTool(
toolName: string,
toolInput: Record<string, unknown>
): Promise<string> {
console.log(`\n> Tool: ${toolName}`);
console.log("Input:", JSON.stringify(toolInput, null, 2));
// In production, this would connect to actual Supabase
if (toolName === "query_postgres") {
return `Query executed: ${toolInput.query}\n(Results would be returned from Supabase)`;
} else if (toolName === "search_vectors") {
return `Semantic search in ${toolInput.table} for "${toolInput.query_text}"\n(Top ${toolInput.top_k || 5} results would be returned)`;
}
return "Tool not found";
}
async function main() {
console.log("=== Jarvis PostgreSQL MCP Agent Demo ===\n");
const userMessage =
"Search for any interactions about budget planning in my memory, then summarize what decisions I made";
console.log("User:", userMessage);
const messages: Anthropic.MessageParam[] = [
{
role: "user",
content: userMessage,
},
];
let continueLoop = true;
while (continueLoop) {
const response = await client.messages.create({
model: "claude-opus-4-1-20250805",
max_tokens: 1024,
tools: tools as Anthropic.Tool[],
messages,
});
console.log("\n--- Claude Response ---");
console.log("Stop Reason:", response.stop_reason);
// Process response content
for (const block of response.content) {
if (block.type === "text") {
console.log("\nText:", block.text);
} else if (block.type === "tool_use") {
console.log(`\n[Tool Use: ${block.name}]`);
const toolResult = await executeTool(
block.name,
block.input as Record<string, unknown>
);
console.log("Result:", toolResult);
// Add tool result to messages
messages.push({
role: "assistant",
content: response.content,
});
messages.push({
role: "user",
content: [
{
type: "tool_result",
tool_use_id: block.id,
content: toolResult,
},
],
});
}
}
// Check if we should continue
if (response.stop_reason === "tool_use") {
// Continue loop to process tool results
continueLoop = true;
} else {
// End of conversation
continueLoop = false;
}
}
console.log("\n=== Demo Complete ===");
}
main().catch(console.error);
