Are you the author? Sign in to claim
๐ Build AI agents that seamlessly combine LLM reasoning with real-world actions via MCP tools โ in just a few lines of
The TypeScript SDK for Multi-Provider AI Agents
Build agents that chain LLM reasoning with MCP tools. Mix OpenAI, Claude, Mistral in one workflow. Parallel execution, branching, loops. Native retries, streaming, and typed errors.
๐ Read the full documentation at volcano.dev โ
๐ค Automatic Tool SelectionLLM automatically picks which MCP tools to call based on your prompt. No manual routing needed. |
๐งฉ Multi-Agent CrewsDefine specialized agents and let the coordinator autonomously delegate tasks. Like automatic tool selection, but for agents. |
๐ฌ Conversational ResultsAsk questions about what your agent did. Use |
๐ง 100s of ModelsOpenAI, Anthropic, Mistral, Bedrock, Vertex, Azure. Switch providers per-step or globally. |
๐ Advanced PatternsParallel execution, branching, loops, sub-agent composition. Enterprise-grade workflow control. |
๐ก StreamingStream tokens in real-time as LLMs generate them. Perfect for chat UIs and SSE endpoints. |
๐ก๏ธ TypeScript-FirstFull type safety with IntelliSense. Catch errors before runtime. |
๐ ObservabilityOpenTelemetry traces and metrics. Export to Jaeger, Prometheus, DataDog, or any OTLP backend. |
โก Production-ReadyBuilt-in retries, timeouts, error handling, and connection pooling. Battle-tested at scale. |
npm install @volcano.dev/agent
That's it! Includes MCP support and all common LLM providers (OpenAI, Anthropic, Mistral, Llama, Vertex).
import { agent, llmOpenAI, mcp } from "@volcano.dev/agent";
const llm = llmOpenAI({
apiKey: process.env.OPENAI_API_KEY!,
model: "gpt-4o-mini"
});
const weather = mcp("http://localhost:8001/mcp");
const tasks = mcp("http://localhost:8002/mcp");
// Agent automatically picks the right tools
const results = await agent({ llm })
.then({
prompt: "What's the weather in Seattle? If it will rain, create a task to bring an umbrella",
mcps: [weather, tasks] // LLM chooses which tools to call
})
.run();
// Ask questions about what happened
const summary = await results.summary(llm);
console.log(summary);
import { agent, llmOpenAI } from "@volcano.dev/agent";
const llm = llmOpenAI({ apiKey: process.env.OPENAI_API_KEY! });
// Define specialized agents
const researcher = agent({ llm, name: 'researcher', description: 'Finds facts and data' })
.then({ prompt: "Research the topic." })
.then({ prompt: "Summarize the research." });
const writer = agent({ llm, name: 'writer', description: 'Creates content' })
.then({ prompt: "Write content." });
// Coordinator autonomously delegates to specialists
const results = await agent({ llm })
.then({
prompt: "Write a blog post about quantum computing",
agents: [researcher, writer] // Coordinator decides when done
})
.run();
// Ask what happened
const post = await results.ask(llm, "Show me the final blog post");
console.log(post);
We welcome contributions! Please see our Contributing Guide for details.
Apache 2.0 - see LICENSE file for details.
Run Claude Code as an MCP server so any agent can delegate coding tasks to it
Browser automation using accessibility snapshots instead of screenshots
Google's universal MCP server supporting PostgreSQL, MySQL, MongoDB, Redis, and 10+ databases
Official GitHub integration for repos, issues, PRs, and CI/CD workflows