Universal LSP MCP Server
Enables language models to access Language Server Protocol functionality across 13 programming languages with automatic detection and zero configuration.
README
Universal LSP MCP Server
One MCP server to rule all Language Servers — automatic language detection, zero-config setup.
Overview
A Model Context Protocol (MCP) server that gives language models access to Language Server Protocol (LSP) functionality across all major programming languages. Unlike existing LSP-MCP servers, this project supports 13 languages out of the box with zero manual configuration — it automatically detects your codebase, selects the right language server, and exposes all LSP operations through a stable, language-independent tool interface.
Supported Languages
| Language | Language Server | Auto-Detected Via |
|---|---|---|
| Python | pyright / pylsp |
pyproject.toml, setup.py |
| TypeScript | typescript-language-server |
tsconfig.json |
| JavaScript | typescript-language-server |
package.json |
| C# | omnisharp |
*.csproj, *.sln |
| Java | vscode-java / jdtls |
pom.xml, build.gradle |
| Go | gopls |
go.mod |
| Rust | rust-analyzer |
Cargo.toml |
| C / C++ | clangd |
*.c, *.cpp, *.h |
| Ruby | solargraph |
Gemfile |
| PHP | intelephense |
composer.json |
| Kotlin | kotlin-language-server |
build.gradle.kts |
| Swift | sourcekit-lsp |
Package.swift |
Features
- 🔍 Automatic Language Detection — Scans project root for language markers (
package.json,Cargo.toml,go.mod, etc.) - 🔄 Auto Language Server Selection — Hardcoded mapping with fallback servers; installs missing LSPs automatically
- 🤖 Hands-Free Initialization — Clients that support MCP Roots auto-initialize on connect —
lsp_initdisappears from the tool list once all servers start cleanly - 💾 Cross-Session Persistence — Initialized workspaces and their ready languages are remembered across server restarts
- 🔁 Graceful Degradation —
lsp_initreappears if a previously-working server starts failing (e.g. language added, binary missing) - 🛠 12 LSP Tools — Definition, references, symbols, diagnostics, rename, code actions, formatting, and more
- 📝 Read & Write Operations — Both inspection and modification of code via LSP
- 🌐 Polyglot Support — Multiple language servers run simultaneously in the same project
- 📋 Hybrid Responses — Human-readable
textfield + raw LSP data inrawfield - 🔌 MCP Stdio Protocol — Works with any MCP-compatible client
- ⚡ Zero Config — Install and run, no per-language setup required
Installation
Important: Install
lsp-mcpon the same machine where your code lives. The language servers it manages need direct filesystem access to your codebase — they cannot work over a remote connection or on a different machine than your source files.
# npm
npm install -g @theupsider/lsp-mcp@latest
# bun
bun install -g @theupsider/lsp-mcp@latest
Quickstart
VS Code / Cursor
Add to your workspace .vscode/mcp.json (recommended — ensures the server runs on the same machine as your code, including SSH remotes, WSL, and Dev Containers):
{
"servers": {
"lsp-mcp": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@theupsider/lsp-mcp@latest"]
}
}
}
Why workspace config? VS Code runs servers defined in
.vscode/mcp.jsonwherever the workspace lives. Servers defined in your user profile always run locally — which breaks LSP when your code is on a remote machine.
CLI
# Run the server (reads from stdin, writes to stdout)
lsp-mcp
The server starts with no active project. Most clients auto-initialize when they support the MCP Roots protocol (VS Code, Copilot, etc.), but lsp_init remains available as a manual override.
For clients without Roots support, the first action the model must take is calling lsp_init:
lsp_init({ root: "/path/to/your/project" })
lsp_init will:
- Scan the root for language markers and start matching language servers (best-effort)
- Disappear from the tool list when you called it explicitly — subsequent calls to any LSP tool trigger lazy server startup for that file's language if no server was detected at init time
- Return health status for all servers that were started eagerly
Optional: pre-warm specific languages (skips detection, faster cold start):
lsp_init({ root: "/path/to/project", languages: ["python", "typescript"] })
Persistence: Once initialized, the workspace configuration (detected languages) is saved to your OS-standard config directory (~/.config/lsp-mcp/ on Linux, ~/Library/Application Support/lsp-mcp/ on macOS, %APPDATA%\lsp-mcp\ on Windows). On subsequent server startups, the MCP server will automatically reconnect using the last-known root, so you rarely need to call lsp_init again.
Re-emergence: If a language server that was previously healthy fails to start (e.g. you added a new language or removed a binary), lsp_init will reappear in the tool list, signaling the model should re-initialize.
Available Tools
Read-Only Tools
| Tool | Description | Key Parameters | Visibility |
|---|---|---|---|
lsp_init |
Initialize server for a project root | root (required), languages (optional string array) |
Conditional — hidden after a successful explicit lsp_init call |
lsp_definition |
Go to definition | file, line, character |
Always |
lsp_references |
Find all references | file, line, character |
Always |
lsp_document_symbols |
List symbols in a file | file |
Always |
lsp_workspace_symbols |
Search symbols across workspace | query (limit: 100–500 results) |
Always |
lsp_diagnostics |
Get errors & warnings | file (scope: file or workspace) |
Always |
lsp_type_definition |
Go to type definition | file, line, character |
Always |
lsp_implementation |
Find implementations | file, line, character |
Always |
lsp_health |
Check status of all LSP servers | (none) | Always |
Write Tools
| Tool | Description | Key Parameters |
|---|---|---|
lsp_rename |
Rename symbol | file, line, character, newName |
lsp_code_action |
Apply / list code actions | file, line, character, apply |
lsp_formatting |
Format document | file |
lsp_range_formatting |
Format code range | file, range |
Configuration
| Environment Variable | Description | Default |
|---|---|---|
LSP_MCP_LOG_LEVEL |
Log level: error, info, debug |
info |
Setup Scripts
Two helper scripts are included for setting up a development environment:
setup-languages-ubuntu24.sh— Installs all language runtimes, compilers, and toolchains on Ubuntu 24.04 (Python, Node.js, Java, Go, Rust, Ruby, PHP, Kotlin, Swift, etc.)setup-lsp.sh— Installs all language servers (pyright, typescript-language-server, omnisharp, jdtls, gopls, rust-analyzer, clangd, solargraph, intelephense, kotlin-language-server)
# 1. Install language runtimes
chmod +x setup-languages-ubuntu24.sh
./setup-languages-ubuntu24.sh
# 2. Install language servers
chmod +x setup-lsp.sh
./setup-lsp.sh
# 3. Reload PATH
source ~/.bashrc
Troubleshooting
Language server not found
If a language server cannot be auto-installed, the server logs a structured error and continues running for other languages. Manually install the missing server:
# Python
pipx install python-lsp-server
# C#
dotnet tool install -g omnisharp-roslyn
# Java
npm install -g vscode-java
# Ruby
gem install solargraph
Server not detecting language
Ensure your project root contains a language marker file (e.g., package.json for TypeScript, Cargo.toml for Rust). The server scans the directory passed to lsp_init for these markers.
High memory usage
Each language server runs as a separate process. For large projects with many languages, consider limiting the workspace or using LSP_MCP_LOG_LEVEL=info to monitor server health.
Architecture
┌─────────────────────────────────────────────┐
│ MCP Client (AI Model) │
└──────────────────┬──────────────────────────┘
│ MCP Protocol (stdio)
▼
┌─────────────────────────────────────────────┐
│ LSP MCP Server (Node.js) │
│ ┌───────────┐ ┌──────────┐ ┌───────────┐ │
│ │ lsp_init │ │ lsp_... │ │ lsp_... │ │
│ └─────┬─────┘ └────┬─────┘ └─────┬─────┘ │
│ └─────────────┼─────────────┘ │
│ ▼ │
│ Language Router & Adapter │
│ (auto-detects language → selects LSP) │
└──────┬──────────┬──────────┬───────────┬─────┘
│ │ │ │
▼ ▼ ▼ ▼
pyright typescript gopls clangd
pylsp lsp rust- ...
analyzer
License
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。