Build MCP-Powered
AI Agents in Minutes

TypeScript SDK for production-ready AI agents. Chain LLMs, connect MCP tools, build complex workflows.

Using Cursor?

Add the official MCP server for Volcano SDK to always use the latest docs to build agents.

Build with MCP

With automatic MCP tool selection to build quicker and smarter agents that actually do things.

Delegate intelligently

Automatically delegate complex tasks to other sub-agents without any manual directions.

Use 1000s of models

Use the right LLM for the job across all the most popular vendors and clouds.

Observe everything

With native Opentelemetry support you have full visibility on everything the agents are doing.

Stop building legacy workflows.
Start building agents.

Volcano SDK autonomously coordinates work across multiple agents unlike legacy workflows where you need to link them together manually.

workflow.ts
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950
import { agent, llmOpenAI, llmAnthropic, llmMistral } 
from "volcano-sdk";

const gpt = llmOpenAI({ apiKey:
process.env.OPENAI_API_KEY! });
const claude = llmAnthropic({ apiKey:
process.env.ANTHROPIC_API_KEY! });
const mistral = llmMistral({ apiKey:
process.env.MISTRAL_API_KEY! });

await agent()

.then({ llm: gpt, prompt: 
"Extract data from report" })
.then({ llm: claude, prompt: 
"Analyze for patterns" })
.then({ llm: mistral, prompt: 
"Write creative summary" })

.run();
Check
Code Editor Background

Everything You Need to Build Production Agents

MCP First, Not Bolted On

Built from the ground up around Model Context Protocol. Automatic tool discovery, connection pooling, and OAuth 2.1 support.

Mix & Match LLMs Per Step

Familiar .then() and .run() syntax. Build complex multi-step

Stop Writing Boilerplate.

Familiar .then() and .run() syntax. Build complex multi-step workflows with patterns you already know. Full TypeScript support.

Retries, Timeouts, Observability

Three retry strategies out of the box. Per-step timeout configuration. OpenTelemetry tracing for every execution. Built for scale.

Stream Tokens & Results

Stream individual tokens in real-time for live UIs. Or stream step results as they complete. Full async generator support.

Run Steps Concurrently

Execute independent steps in parallel. Build complex workflows with branching, loops, and sub-agent composition.

Stop Writing Boilerplate. Start Building Agents.

Other frameworks need 100+ lines to build an agent. With Volcano, you need 9.

Others

100+ lines of boilerplate

Manual tool schema handling

Provider-specific complexity

DIY context management

Custom retry logic needed

No streaming support

Zero observability

Volcano

100+ lines of boilerplate

Manual tool schema handling

Provider-specific complexity

DIY context management

Custom retry logic needed

No streaming support

Zero observability

Volcano

Frequently
Asked Questions