Claude Agent SDK - Danube

Install

npm install danube @anthropic-ai/claude-agent-sdk
export DANUBE_API_KEY=...

Option 1: the hosted MCP server (no code in the loop)

Danube’s MCP server at https://mcp.danubeai.com/mcp already exposes search, describe, execute and the rest. Register it and allow the tools you want:

import { query } from '@anthropic-ai/claude-agent-sdk';
import { danubeMcpServer, danubeMcpToolNames } from 'danube/claude-agent-sdk';

for await (const message of query({
  prompt: 'Post "deploy finished" to #ops on Slack',
  options: {
    mcpServers: { danube: danubeMcpServer({ apiKey: process.env.DANUBE_API_KEY!, groups: ['core'] }) },
    allowedTools: danubeMcpToolNames(['search_tools', 'describe_tool', 'execute_tool']),
  },
})) {
  if (message.type === 'result') console.log(message.result);
}

groups: ['core'] asks the server for its small core tool list (see MCP tool groups); omit it for everything.

Option 2: in-process tools

The same three tools as every other adapter, served through the SDK’s own MCP server. The SDK’s tool helper and zod are passed in, so danube adds no dependency:

import { query, createSdkMcpServer, tool } from '@anthropic-ai/claude-agent-sdk';
import { z } from 'zod';
import { DanubeClient } from 'danube';
import { danubeSdkTools } from 'danube/claude-agent-sdk';

const danube = createSdkMcpServer({
  name: 'danube',
  tools: danubeSdkTools(new DanubeClient(), { tool, z }),
});

query({ prompt: 'Top Hacker News stories?', options: { mcpServers: { danube } } });

Options for danubeSdkTools: serviceId, readyOnly, maxResults.

tool	purpose
`danube_search_tools`	find tools across the catalog
`danube_describe_tool`	schema, tips, readiness before a first call
`danube_execute_tool`	run it; errors say what to do next