Universal LSP MCP Server

One MCP server to rule all Language Servers — automatic language detection, zero-config setup.

License: Apache-2.0 npm version Node.js Platform

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_init disappears from the tool list once all servers start cleanly
  • 💾 Cross-Session Persistence — Initialized workspaces and their ready languages are remembered across server restarts
  • 🔁 Graceful Degradationlsp_init reappears 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 text field + raw LSP data in raw field
  • 🔌 MCP Stdio Protocol — Works with any MCP-compatible client
  • ⚡ Zero Config — Install and run, no per-language setup required

Installation

Important: Install lsp-mcp on 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.json wherever 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:

  1. Scan the root for language markers and start matching language servers (best-effort)
  2. 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
  3. 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

Apache-2.0