What it does

agent-lsp is an MCP server that sits between AI coding agents and the language servers editors already use, such as gopls, rust-analyzer and pyright. It is not a language server itself; it manages them and exposes their code intelligence to agents as MCP tools. The pitch is that one call can replace many: a blast-radius tool returns an exported symbol's callers in a single step, and speculative-editing tools let an agent preview an edit's effect on diagnostics before anything is written to disk. The project also ships workflow skills, such as refactor, safe-edit and verify, that chain these tools in a fixed order, and it keeps language servers warm between sessions. It is a single Go binary with install options including a shell script, Homebrew, pip, npm, winget, Scoop and Docker. The site lists setup for Claude Code, Cursor, Windsurf, Gemini CLI, Continue and other MCP clients, and says CI runs real language servers for 32 languages.

Capabilities

Over 60 MCP tools for references, call hierarchy, diagnostics, renames and more
Speculative editing: preview an edit's diagnostic impact before writing to disk
Workflow skills that chain impact analysis, edit, verify and test steps
Persistent runtime that keeps language servers indexed across sessions
One process routes files to the right language server across 32 CI-tested languages
Runs over stdio or HTTP as a single Go binary

When to use it

Automating refactors across large or multi-language codebases with an AI agent
Checking what a rename or edit will break before it touches any file
CI or scripted tooling that needs programmatic code intelligence