mcpstore

Why mcpstore

mcpstore is built with Rust for high performance and low overhead. Its architecture mirrors Kubernetes-style control plane / data plane: the control plane handles orchestration, while the data plane focuses on invocation. On resource-constrained hosts you can attach only the data plane and call already-shared MCP services with zero local maintenance. SDKs are available for embedding into applications and Agents.

Quick Start

npm install -g ip2a@mcpstore

or

curl -fsSL https://raw.githubusercontent.com/ip2a/mcpstore/main/install.sh | bash

Prefer code integration? → Use via SDK

CLI Usage

Manage MCP services and runtime state from the CLI. See Quick Start above for installation.

Add, connect, restart, or remove a service:

mcpstore add wiki https://example.com/mcp
mcpstore list
mcpstore connect wiki
mcpstore remove wiki

Inspect local config, or open the Web UI:

mcpstore config show
mcpstore web

Expose an HTTP API:

mcpstore api --host 127.0.0.1 --port 1820

Desktop App (Beta)

The GUI is the control plane: manage MCP services and tools, inspect parameters, and try invocations. Currently in Beta — download from GitHub Releases.

SDK Usage

Integrate mcpstore into your application via the Python SDK or Rust Lib.

Install

pip install mcpstore

Quick Start

Initialize a store:

from mcpstore import MCPStore

store = MCPStore.setup_store()

Use store.for_store() to manage MCP services and tools in the global scope.

Add your first service

store.for_store().add_service({
    "mcpServers": {
        "mcpstore_wiki": {
            "url": "https://example.com/mcp"
        }
    }
})
store.for_store().wait_service("mcpstore_wiki")

Convert to LangChain tools and use them

from langchain.agents import create_agent
from langchain_openai import ChatOpenAI

tools = store.for_store().for_langchain().list_tools()

llm = ChatOpenAI(
    temperature=0,
    model="your-model",
    api_key="sk-*****",
    base_url="https://api.xxx.com",
)
agent = create_agent(model=llm, tools=tools, system_prompt="You are an assistant")
events = agent.invoke({
    "messages": [{"role": "user", "content": "How do I add a service in mcpstore?"}]
})
print(events)

Group by Agent

Use for_agent(agent_id) to create an independent scope per Agent:

store.for_agent("agent1").add_service({
    "name": "mcpstore_wiki",
    "url": "https://example.com/mcp",
})

store.for_agent("agent2").add_service({
    "name": "github",
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-github"]
    },
})

agent1_tools = store.for_agent("agent1").list_tools()
agent2_tools = store.for_agent("agent2").list_tools()

Install

cargo add mcpstore

Quick Start

use std::time::Duration;

use mcpstore::{McpConfig, MCPStore, Result, ServiceTarget};

#[tokio::main]
async fn main() -> Result<()> {
    let store = MCPStore::setup(None)?;

    store
        .for_store()
        .add_service(McpConfig::from_json_str(
            r#"{"name":"mcpstore_wiki","url":"https://example.com/mcp"}"#,
        )?)
        .await?;

    store
        .for_store()
        .wait_service(
            ServiceTarget::ServiceName("mcpstore_wiki"),
            Duration::from_secs(30),
        )
        .await?;

    let tools = store.for_store().list_tools().await?;
    println!("loaded tools: {}", tools.len());
    Ok(())
}

Star History

Issues and suggestions are welcome via GitHub Issues.