MCP Catalogs
Home

agent-lsp

by blackwell-systems·26·Score 45

Stateful MCP server bridging language servers with AI agents, offering type-aware navigation, blast-radius analysis, and pre-verified edits across 30+ languages.

developer-toolsai-llm
1
Forks
1
Open issues
this month
Last commit
2d ago
Indexed

Overview

agent-lsp is a stateful runtime that bridges language servers and AI agents, providing 65+ tools for type-aware code navigation, blast-radius analysis, and pre-verified edits. It solves the problem of AI agents making incorrect code changes by giving them full visibility into code structure, dependencies, and potential impacts. The solution indexes workspaces once, keeps the index warm, and adds a skill layer that encodes correct multi-step operations. Unlike other MCP-LSP implementations, it has been CI-verified against 30 real language servers and offers unique features like speculative execution and phase enforcement.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

you:Large-scale refactoring with blast-radius analysis to understand impact before making changes
you:Type-safe navigation across codebases to find implementations, references, and call hierarchies
you:Speculative editing to preview changes in memory before writing to disk
you:What languages are supported?
you:How does agent-lsp differ from other MCP-LSP implementations?

When to choose this

Choose agent-lsp when working with large codebases where type-aware navigation, impact analysis, and safe refactoring are critical. It's ideal for teams that need to maintain code quality across multiple programming languages.

When NOT to choose this

Avoid agent-lsp if you're working on small projects where the overhead of a persistent daemon isn't justified, or if you need support for languages not in its 30-language verified list.

Tools this server exposes

12 tools extracted from the README
  • blast_radius

    Find all call sites of a symbol with test/non-test partitioning

  • go_to_implementation

    Find all concrete implementations of an interface

  • preview_edit

    Preview the diagnostic impact of an edit without writing to disk

  • simulate_chain

    Evaluate a sequence of dependent edits and find where the first error occurs

  • find_references

    Find all references to a symbol in the codebase

  • get_diagnostics

    Get current diagnostic issues (errors, warnings) in a file or workspace

  • replace_symbol_body

    Replace the implementation of a symbol while preserving its signature

  • safe_apply_edit

    Preview edits and apply to disk only if no new errors are introduced

  • find_callers

    Find functions that call a specific symbol

  • get_definition

    Go to the definition location of a symbol

  • get_hover

    Get hover information (type, documentation) for a symbol

  • safe_delete_symbol

    Delete a symbol safely after finding all its references

Comparable tools

mcp-server-lspmcp-xbaseshell-mcpgithub-mcp

Installation

Install agent-lsp as a Go binary:

go install github.com/blackwell-systems/agent-lsp@latest

For Claude Desktop, add to .mcp.json:

{
  "mcpServers": {
    "agent-lsp": {
      "command": "agent-lsp",
      "args": ["stdio"]
    }
  }
}

Docker images are available for each language (go, typescript, python, etc.)

FAQ

What languages are supported?
agent-lsp supports 30+ CI-verified languages including Go, TypeScript, Python, Rust, Java, C/C++, C#, Ruby, PHP, Kotlin, Swift, Scala, Zig, Lua, Elixir, and more. Each language server is actually tested in CI.
How does agent-lsp differ from other MCP-LSP implementations?
Unlike other implementations, agent-lsp is a stateful runtime that keeps indexes warm, offers CI-verified language server support, provides speculative execution capabilities, enforces workflow phases, and includes features like concurrency analysis and token savings.

Compare agent-lsp with

GitHub →

Last updated · Auto-generated from public README + GitHub signals.