M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
V

Volcano Agent SDK

Community
Archived
Framework / SDK
Build AI agents that combine LLM reasoning with MCP tool calls in a few lines of TypeScript

This is a framework for building MCP servers or agents, not a server you connect a client to.

How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #370 of 438
Stars
★ 406 Popular
Runtime
Node.js
Credentials
API key / credential required
License
Other / unspecified
Last commit
⚠
38FMRS · D

Volcano Agent SDK is a multi-provider AI agent orchestration framework for TypeScript developers that connects to external MCP servers for automatic tool selection and multi-agent coordination, but it is not itself an MCP server exposing tools, and the repository provides no server manifest or fixed tool list.

Strongest · Maintenance 11/20 Weakest · Reliability 4/20

Reliability
4/20
Security and permissions
8/20
Maintenance
11/20
Documentation
9/20
Setup experience
6/20
Why each score
Reliability 4/20
The server manifest is marked "not cached," and no tool list, initialization handshake trace, or test files are present in the evidence provided. The README indicates this project is actually an agent SDK/MCP client (it connects out to external MCP servers via mcp("http://localhost:8001/mcp")) rather than an MCP server exposing its own tools, so there is no way to verify declared tools against real behavior or confirm the MCP init/tool-list handshake. A CI badge is shown but no workflow file or test content was supplied to demonstrate coverage of key paths. Per the static calibration rule, missing evidence is a deduction rather than a best-case assumption, so this scores low.
Security and permissions 8/20
Example code passes API keys via environment variables (e.g., OPENAI_API_KEY) with no hardcoded secrets or credential-leak evidence found. However, the README describes an LLM "automatically" selecting and invoking MCP tools and a coordinator that "autonomously delegates tasks" until done, with no mention anywhere of scoped permissions, human confirmation for consequential actions, or disclosure of what data flows to which LLM provider or MCP endpoint. These are visible but unaddressed risks — not a red line, since there is no evidence of malware, credential theft, or irreversible delete/pay/deploy defaults — but the absence of least-privilege framing or confirmation steps warrants a clear deduction.
Maintenance 11/20
The repo shows a CI badge, a published npm package, 401 stars, only 10 open issues, and is not archived, suggesting some activity. However, GitHub's license field reads "NOASSERTION" while the README claims Apache 2.0 with a LICENSE file — a governance inconsistency that needs clarification. No verifiable evidence of release cadence, recent commit activity, dependency-update practices, or a security-response channel (e.g., SECURITY.md) was provided, so scoring stays conservative given the limited information.
Documentation 9/20
The README offers install instructions, a quick-start example, and a feature overview, which is reasonably readable, but the core documentation lives on an external site (volcano.dev) that cannot be statically verified from the repository. Details on MCP tool parameters, authentication flows, rate limits, cost, and troubleshooting are not present in the provided material, leaving hidden assumptions about key usage details, which caps the score.
Setup experience 6/20
Installation is simple (npm install plus a few lines of example code), but this project is fundamentally a library for writing agent code that connects to other MCP servers, not an MCP server that can be configured directly into mainstream clients like Claude Desktop or Cursor — so no relevant client-configuration example exists or applies. No committed CI workflow content or tests demonstrating a verified install/run path were supplied, so per the static calibration rule setup cannot exceed 15, and it is marked down further given how little verifiable setup evidence is available.

Static review · not runListed 2026-08-26

Read the FMRS scoring method →

Fit and risk

What it can accessUses the network

Best for

  • Backend/platform engineers already using TypeScript who need to orchestrate LLM calls and multiple MCP servers within their own application
  • Teams building multi-agent delegation workflows (research, writing, task handling, etc.) with autonomous coordination
  • Teams that need to mix several LLM providers in one workflow with built-in retries, streaming, and tracing

Not for

  • Users who want an out-of-the-box MCP server that exposes specific business tools (this project provides no tools itself — it requires connecting to external MCP servers)
  • Developers working in Python or other non-JavaScript/TypeScript stacks
  • Non-technical users hoping to add it via a Claude Desktop-style configuration file

Required permissions

  • Requires developer-supplied API keys for the LLM provider(s) in use (e.g., OPENAI_API_KEY)
  • Requires network access to connect to whatever MCP server endpoints the developer deploys (e.g., http://localhost:8001/mcp)
  • Because it supports automatic tool selection, it implicitly gains whatever permissions the connected MCP servers grant to their tools

Risks and side effects

  • Automatic tool selection means the LLM decides which tools to invoke based on the prompt; if connected to MCP servers with real side effects (creating tasks, sending requests), there is a risk of unintended actions from prompt injection or model misjudgment
  • In multi-agent delegation mode, the coordinator autonomously decides when the workflow is done, which could lead to unexpected repeated calls or resource use if not designed carefully
  • The README does not describe specific authentication, sandboxing, or permission-isolation mechanisms — the security boundary of connected MCP servers is the developer's responsibility

Setup

Before you start

Runtime：Node.js

OPENAI_API_KEY requiredsecret API key for OpenAI models, required by llmOpenAI(); create it in the API keys section at platform.openai.com.

Install via npm: npm install @volcano.dev/agent. Then, in TypeScript code, import agent, llmOpenAI (or another provider helper), and mcp; initialize the model with your LLM API key, point mcp("http://host:port/mcp") at an already-running MCP server, chain .then() steps, and call .run(). The SDK does not require or provide a JSON configuration file for clients like Claude Desktop, since it is a library embedded in your own TypeScript application rather than a standalone MCP server process.

Check that it works

After npm install @volcano.dev/agent, run the README Hello World example pointing mcp() at http://localhost:8001/mcp and prompt 'What's the weather in Seattle?' — if the LLM automatically calls the MCP tools and returns results, the SDK is working.

Troubleshooting

  1. Verify the LLM provider API key environment variables are set correctly
  2. Confirm the server URL passed to mcp() is reachable and the target MCP server is running and listening on the expected port
  3. Consult the official docs at volcano.dev/docs for MCP tool connection pooling and OAuth authentication sections when tools aren't being selected or invoked as expected
  4. Check that the @volcano.dev/agent npm package is up to date, or review the repository's CI status for known build issues

Things to try

Once connected, you can ask your AI assistant things like:

  • Build an agent that calls a local MCP weather server with agent() and mcp()
  • Swap llmOpenAI for llmAnthropic and run the same multi-step workflow
  • Delegate a writing task to a coordinator using agents: [researcher, writer]
  • Ask what the agent did using results.summary(llm)

Use cases

Orchestrating an LLM agent that automatically selects and calls tools across multiple deployed MCP servers based on a natural-language prompt
Building multi-agent pipelines where a coordinator agent autonomously delegates work to specialized sub-agents (e.g., a researcher and a writer)
Running production workflows that mix or switch between multiple LLM providers (OpenAI, Anthropic, Mistral, etc.) with built-in OpenTelemetry tracing

Overview

Volcano Agent SDK is a TypeScript SDK from Kong for building AI agents that chain LLM reasoning with real-world actions performed through MCP tools. It is important to note this project itself is a client-side orchestration framework, not an MCP server exposing tools — developers connect to external MCP servers via mcp(url) and let the LLM automatically pick which of those servers' tools to call. It supports mixing multiple LLM providers (OpenAI, Anthropic, Mistral, and others), and offers parallel execution, branching, loops, multi-agent coordination (a coordinator agent autonomously delegates tasks to specialized sub-agents), streaming output, OpenTelemetry observability, and built-in retries/timeouts. No server manifest or fixed tool list is provided in the repository, since the available tools depend entirely on whichever external MCP servers a developer connects.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 1d3e13a3aad4 Data synced 2026-10-19 Read the FMRS scoring method