open MCP Client

Use This MCP client To

Check that a tool I just wrote returns what I expect Chat with a model and watch which of my tools it picks Test several servers at once from a single window Save a debugging session and pick it up tomorrow Compare how different models handle the same tool

README

OpenMCP: All you need for MCP Development

English | 中文

🫱 Official Documentation

OpenMCP QQ Group

OpenMCP Discord Channel

OpenMCP

An all-in-one vscode/trae/cursor plugin for MCP server debugging.

👆 Full Video

openmcpent

Integrated Inspector + MCP client basic functions, combining development and testing into one.

Test mcp tools, prompts and resources with a variety of tools.

Tested tools can be placed in the "Interactive Testing" module for large model interaction testing.

Complete project-level management panel for easier MCP project management at both project and global levels.

Supports multiple large models: DeepSeek, OpenAI, Qwen, Gemini, Grok, Mistral, MiniMax, Groq, Perplexity, Kimi, Ollama, OpenRouter and more.

Support XML mode and customized options for your tool selection.

openmcp-sdk

once everything is tested and verified in openmcpent, you can deploy your mcp as an agent app with openmcp-sdk fastly and easily:

npm install openmcp-sdk

then deploy your agent with just lines of codes

import { OmAgent } from 'openmcp-sdk/service/sdk';

// create Agent
const agent = new OmAgent();
    
// Load configuration, which can be automatically generated after debugging with openmcp client
agent.loadMcpConfig('./mcpconfig.json');

// Read the debugged prompt
const prompt = await agent.getPrompt('hacknews', { topn: '5' });    

// Execute the task
const res = await agent.ainvoke({ messages: prompt });

console.log('⚙️ Agent Response', res);

output

[2025/6/20 20:47:31] 🚀 [crawl4ai-mcp] 1.9.1 connected
[2025/6/20 20:47:35] 🤖 Agent wants to use these tools get_web_markdown
[2025/6/20 20:47:35] 🔧 using tool get_web_markdown
[2025/6/20 20:47:39] ✓  use tools success
[2025/6/20 20:47:46] 🤖 Agent wants to use these tools get_web_markdown, get_web_markdown, get_web_markdown
[2025/6/20 20:47:46] 🔧 using tool get_web_markdown
[2025/6/20 20:47:48] ✓  use tools success
[2025/6/20 20:47:48] 🔧 using tool get_web_markdown
[2025/6/20 20:47:54] ✓  use tools success
[2025/6/20 20:47:54] 🔧 using tool get_web_markdown
[2025/6/20 20:47:57] ✓  use tools success

⚙️ Agent Response
⌨️ Today's Tech Article Roundup

📌 How to Detect or Observe Passing Gravitational Waves?
Summary: This article explores the physics of gravitational waves, explaining their effects on space-time and how humans might perceive or observe this cosmic phenomenon.
Author: ynoxinul
Posted: 2 hours ago
Link: https://physics.stackexchange.com/questions/338912/how-would-a-passing-gravitational-wave-look-or-feel

📌 Learn Makefile Tutorial
Summary: A comprehensive Makefile tutorial for beginners and advanced users, covering basic syntax, variables, automatic rules, and advanced features to help developers manage project builds efficiently.
Author: dsego
Posted: 4 hours ago
Link: https://makefiletutorial.com/

📌 Hurl: Run and Test HTTP Requests in Plain Text
Summary: Hurl is a command-line tool that allows defining and executing HTTP requests in plain text format, ideal for data fetching and HTTP session testing. It supports chained requests, value capture, and response queries, making it perfect for testing REST, SOAP, and GraphQL APIs.
Author: flykespice
Posted: 8 hours ago
Link: https://github.com/Orange-OpenSource/hurl

🔗 Document of openmcp-sdk

Want to become a contributor ?

Click here to learn how to make contribution to OpenMCP.

Join Our Community

Email: zhelonghuang@qq.com
QQ: 782833642
Wechat: contact lstmkirigaya
Discord: https://discord.gg/SKTZRf6NzU

TODO

Feature Roadmap

Module	Feature	Priority	Status	Fix Priority
`all`	Complete basic infrastructure	`Full Version`	100%	`Done`
`render`	Support cost analysis in chat mode	`Iteration`	100%	`Done`
`ext`	Support basic MCP project management	`Iteration`	100%	`P0`
`service`	Support custom OpenAI-compatible large model integration	`Full Version`	100%	`Done`
`service`	Support custom protocol large model integration	`MVP`	0%	`P1`
`all`	Support debugging multiple MCP Servers simultaneously	`MVP`	100%	`P0`
`all`	Support online verification via large models	`Iteration`	100%	`Done`
`all`	Support saving user's server debugging work	`Iteration`	100%	`Done`
`render`	High-risk operation permission confirmation	`MVP`	0%	`P1`
`service`	Hot update for connected MCP servers	`MVP`	0%	`P1`
`all`	System prompt management module	`Iteration`	100%	`Done`
`service`	Tool-wise logging system	`MVP`	0%	`P1`
`service`	MCP security checks (prevent prompt injection, etc.)	`MVP`	0%	`P1`
`service`	Built-in OCR for character recognition	`Iteration`	100%	`Done`

Project Concept

OpenMCP adopts a layered modular design. By assembling different modules, it can be implemented in different modes on different platforms.

flowchart TD
    subgraph OpenMCP Core Components
        renderer[Renderer]
        openmcpservice[OpenMCPService]
    end

    subgraph OpenMCP_Web["OpenMCP Web"]
        renderer
        openmcpservice
        nginx[Nginx]
    end

    subgraph OpenMCP_Plugin["OpenMCP Plugin"]
        renderer
        openmcpservice
        vscode[VSCode Plugin Code]
    end

    subgraph OpenMCP_App["OpenMCP App"]
        renderer
        openmcpservice
        electron[Electron Code]
    end

    subgraph QQBot["OpenMCP-based QQ Bot"]
        lagrange[Lagrange.OneBot]
        openmcpservice
    end

    %% Dependencies
    OpenMCP_Web -->|Frontend Rendering| renderer
    OpenMCP_Web -->|Backend Service| openmcpservice
    OpenMCP_Web -->|Reverse Proxy| nginx

    OpenMCP_Plugin -->|UI Interface| renderer
    OpenMCP_Plugin -->|Core Logic| openmcpservice
    OpenMCP_Plugin -->|IDE Integration| vscode

    OpenMCP_App -->|Frontend UI| renderer
    OpenMCP_App -->|Local Service| openmcpservice
    OpenMCP_App -->|Desktop Packaging| electron

    QQBot -->|Protocol Adaptation| lagrange
    QQBot -->|Business Logic| openmcpservice

OpenMCP CLI

Quickly setup and run OpenMCP development environment with a single command:

# Install CLI globally
npm install -g @agent-ruler/openmcp

# Create a new project
openmcp init my-project
cd my-project

# Start development servers (service + renderer)
openmcp dev

CLI Commands

Command	Description
`openmcp init <project-name>`	Initialize a new OpenMCP project
`openmcp dev`	Start development mode (backend + frontend)
`openmcp dev --service-only`	Start only the backend service
`openmcp dev --renderer-only`	Start only the frontend renderer
`openmcp start`	Start production mode
`openmcp update`	Update to the latest version

📖 Full CLI Documentation

Development

renderer : Frontend UI definitions
service : Test components for renderer , including a simple forwarding layer
src : VSCode plugin definitions

Renderer & Service Development

flowchart LR
D[renderer] <--> A[Dev Server] 
<--ws--> B[service]
B <--mcp--> m(MCP Server)

Project setup:

npm run setup

Start dev server:

npm run serve

Extension Development

flowchart LR
D[renderer] <--> A[extention.ts] <--> B[service]
B <--mcp--> m(MCP Server)

Build for deployment:

npm run build

build vscode extension:

npm run build:plugin

Then just press F5, いただきます (Let's begin)

CI Pipeline

✅ npm run build ✅ npm run build:task-loop ✅ openmcpent UT ✅ openmcp-sdk UT ✅ vscode extension UT

openmcp-client FAQ

Which editors does it work in?

VS Code, and also Cursor and Trae, which are built on the same foundation.

Do I need a key?

Yes, you bring your own model key. It supports DeepSeek, OpenAI, Gemini, Grok, Qwen, Kimi, Mistral, Groq, Perplexity, and OpenRouter, plus local models through Ollama.

Is this for people who just want to use MCP servers?

Not really. It is a workbench for building and debugging your own servers, rather than an everyday assistant.

Can I use this to work out why a tool is failing?

Yes. You can call the tool directly and read the raw reply, then try the same request through a model to see where things go wrong.

Does it cost anything?

The extension itself is free and open source. You only pay whatever your chosen model provider charges.

Can it handle more than one server at a time?

Yes. Connecting to several at once is supported, and there is a management panel for both project-level and global setups.