Because LLMs can't count lines: an LSP bridge for Claude Code

What it does cclsp acts as a bridge between Model Context Protocol clients like Claude Code and standard Language Server Protocol servers. It exposes LSP capabilities—definition lookup, reference finding, symbol renaming, and diagnostics—as MCP tools that an AI agent can invoke. You configure it with a JSON file pointing to whatever language servers you have installed for your stack.

The interesting bit The README is unusually honest about the core problem: LLM coding assistants “often struggle with providing accurate line/column numbers,” which makes direct LSP integration fragile. cclsp solves this by intelligently trying multiple position combinations until it resolves a symbol, effectively compensating for the AI’s shaky spatial reasoning.

Key highlights

• Exposes six MCP tools: find_definition, find_references, rename_symbol, rename_symbol_strict, get_diagnostics, and restart_server
• Works with any LSP-compatible server; the README lists configurations for TypeScript, Python, Go, Rust, C/C++, Java, Ruby, PHP, C#, and Swift
• Includes an interactive setup wizard that auto-detects project languages and suggests matching language servers
• rename_symbol applies changes across all affected files and creates .bak backups automatically; if the symbol is ambiguous, it returns candidates and falls back to rename_symbol_strict

Caveats

• You must install and maintain language servers separately; cclsp is strictly the glue layer
• Symbol renaming can return multiple candidates when the AI’s guess is ambiguous, requiring manual disambiguation via rename_symbol_strict
• Only as useful as the underlying LSP servers you configure; there is no built-in language intelligence beyond the adapter logic

Verdict Useful if you want IDE-grade code navigation inside Claude Code without switching contexts. If you already live in VS Code or JetBrains, this adds little.