markus41/plugins/claude-code-expert/archive/v7.6.0/skills/lsp-integration

LSP Integration — Phase 7 of /cc-setup

LSP (Language Server Protocol) servers power real-time diagnostics, type checking, and formatting. Claude Code integrates with LSP through PostToolUse hooks to catch errors immediately after writing files.

18-Server Detection Matrix

| Language | LSP Server | Install Command | Detect Signal | |----------|-----------|-----------------|---------------| | TypeScript/JS | typescript-language-server | npm i -g typescript-language-server typescript | tsconfig.json, package.json | | Python (strict) | pyright | npm i -g pyright | pyproject.toml, requirements.txt | | Python (general) | pylsp | pip install python-lsp-server | .py files | | Go | gopls | go install golang.org/x/tools/gopls@latest | go.mod | | Rust | rust-analyzer | rustup component add rust-analyzer | Cargo.toml | | Java | jdtls | Eclipse JDT LS (via Mason/LspInstall) | pom.xml, build.gradle | | C# | omnisharp | dotnet tool install -g csharp-ls | *.csproj, *.sln | | Ruby | solargraph | gem install solargraph | Gemfile | | PHP | intelephense | npm i -g intelephense | composer.json | | Elixir | elixir-ls | mix archive.install hex elixir_ls | mix.exs | | Svelte | svelte-language-server | npm i -g svelte-language-server | svelte.config.* | | Vue | vue-language-server | npm i -g @vue/language-server | nuxt.config.* | | Tailwind CSS | tailwindcss-language-server | npm i -g @tailwindcss/language-server | tailwind.config.* | | GraphQL | graphql-language-service-cli | npm i -g graphql-language-service-cli | *.graphql, schema.gql | | Prisma | prisma-language-server | npm i -g @prisma/language-server | prisma/ directory | | YAML | yaml-language-server | npm i -g yaml-language-server | .yaml, .yml files | | Dockerfile | dockerfile-language-server | npm i -g dockerfile-language-server-nodejs | Dockerfile | | Bash | bash-language-server | npm i -g bash-language-server | .sh files, .claude/hooks/ |

Detection Script

Run during Phase 1 of /cc-setup to identify which LSPs to recommend:

#!/usr/bin/env bash
# Detect and report LSP recommendations for current project
check_lsp() {
  local name="$1" signal="$2" install="$3"
  if ls $signal 2>/dev/null | head -1 | grep -q .; then
    if command -v "$(echo "$name" | cut -d/ -f1)" &>/dev/null; then
      echo "  OK:      $name"
    else
      echo "  MISSING: $name — install: $install"
    fi
  fi
}

echo "=== LSP Detection ==="
check_lsp "typescript-language-server" "tsconfig.json package.json" "npm i -g typescript-language-server typescript"
check_lsp "pyright" "pyproject.toml requirements.txt" "npm i -g pyright"
check_lsp "gopls" "go.mod" "go install golang.org/x/tools/gopls@latest"
check_lsp "rust-analyzer" "Cargo.toml" "rustup component add rust-analyzer"
check_lsp "solargraph" "Gemfile" "gem install solargraph"
check_lsp "svelte-language-server" "svelte.config.*" "npm i -g svelte-language-server"
check_lsp "tailwindcss-language-server" "tailwind.config.*" "npm i -g @tailwindcss/language-server"
check_lsp "prisma-language-server" "prisma/" "npm i -g @prisma/language-server"
check_lsp "bash-language-server" ".claude/hooks/" "npm i -g bash-language-server"

Claude Code Hook Integration

Claude Code doesn't use LSP directly, but you can use PostToolUse hooks to run LSP-powered diagnostics after file writes.

TypeScript — PostToolUse diagnostics hook

#!/usr/bin/env bash
# .claude/hooks/auto-typecheck.sh
# Registered on: PostToolUse (Write|Edit matcher on *.ts, *.tsx)
INPUT=$(head -c 65536)
FILE=$(printf '%s' "$INPUT" | jq -r '.tool_input.file_path // .tool_input.path // ""')

if [[ "$FILE" != *.ts ]] && [[ "$FILE" != *.tsx ]]; then
  echo '{"decision": "approve"}'
  exit 0
fi

# Run whole-project check filtered to changed file — single-file mode ignores tsconfig.json
ERRORS=$(npx tsc --noEmit 2>&1 | grep -F "$(basename "$FILE")" | head -5)
if [ -n "$ERRORS" ]; then
  echo "TypeScript errors in $(basename "$FILE"):" >&2
  echo "$ERRORS" >&2
fi

echo '{"decision": "approve"}'

Python — pyright diagnostics hook

#!/usr/bin/env bash
# Registered on: PostToolUse (Write|Edit matcher on *.py)
INPUT=$(head -c 65536)
FILE=$(printf '%s' "$INPUT" | jq -r '.tool_input.file_path // ""')

if [[ "$FILE" != *.py ]]; then
  echo '{"decision": "approve"}'
  exit 0
fi

ERRORS=$(pyright "$FILE" 2>&1 | grep -E "error:" | head -5)
if [ -n "$ERRORS" ]; then
  echo "Pyright errors:" >&2
  echo "$ERRORS" >&2
fi

echo '{"decision": "approve"}'

Rust — cargo check hook

#!/usr/bin/env bash
# Registered on: PostToolUse (Write|Edit matcher on *.rs)
INPUT=$(head -c 65536)
FILE=$(printf '%s' "$INPUT" | jq -r '.tool_input.file_path // ""')

if [[ "$FILE" != *.rs ]]; then
  echo '{"decision": "approve"}'
  exit 0
fi

cargo check 2>&1 | grep -E "^error" | head -10 >&2
echo '{"decision": "approve"}'

settings.json Hook Registration

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Write|Edit",
        "hooks": [
          {
            "type": "command",
            "command": "bash .claude/hooks/auto-typecheck.sh"
          }
        ]
      }
    ]
  }
}

Why LSP Integration Matters

• Immediate feedback: Errors surface right after Claude writes a file, not at build time
• Type-safe edits: TypeScript errors caught before Claude attempts to run tests
• Batch diagnostics: Run tsc --noEmit across the whole project to catch cross-file breakage
• Cost savings: Catching errors early avoids multiple round-trips of write → test → fix

LSP vs Built-in IDE Diagnostics

Claude Code has a built-in LSP tool (via mcp__ide__getDiagnostics) when running in VS Code or JetBrains IDE extensions. The hook approach above is for CLI usage where no IDE is attached.

| Context | Diagnostics Source | |---------|-------------------| | VS Code extension | mcp__ide__getDiagnostics (built-in) | | JetBrains extension | mcp__ide__getDiagnostics (built-in) | | CLI (terminal) | PostToolUse hook → tsc --noEmit / pyright | | CI/CD headless | PostToolUse hook → LSP CLI tools |

Cost Note

LSP installation is one-time. The diagnostics hooks add ~100ms per file write, which is negligible vs the cost of a missed type error causing a failed test run (which requires another full turn).