# LSAI + xmp4

> Two MCP servers that give AI coding agents compiler-grade code intelligence. LSAI analyzes your local code in real time via LSP. xmp4 provides semantic navigation of 900+ pre-indexed OSS libraries via SCIP. Both free, both open protocol (MCP).

## LSAI — your code, live

LSAI runs locally on your machine and connects to LSP servers (Roslyn, jdtls, Pyright, rust-analyzer, etc.) to give your AI agent real-time compiler-grade intelligence on the codebase you are actively developing.

- 14 MCP tools: search, info, usages, callers, callees, hierarchy, impact, rename, outline, diagnostics, deps, source, file_refs, context
- 9 languages: C#, TypeScript, Python, Java, Rust, PHP, Go, JavaScript, C/C++
- Runs 100% locally — your code never leaves your machine
- Install: `curl -fsSL https://github.com/0ics-srls/Zerox.Lsai.Public/releases/latest/download/install.sh | bash`
- Docs: https://github.com/0ics-srls/Zerox.Lsai.Public

## xmp4 — every library, pre-indexed

xmp4 is a hosted MCP server with 900+ pre-indexed OSS libraries across 12 languages. It answers symbol-level questions (real source, typed callers, usages, tests, hierarchies) in a single tool call.

- 17 MCP tools: projects, search, info, source, usages, callers, callees, hierarchy, outline, deps, tests_for, view, grep, symbol_at, server, guide
- 12 languages: C#, TypeScript, Python, Java, Rust, PHP, Go, JavaScript, C/C++, Dart, Ruby, Scala
- 900+ OSS libraries, 15,000+ projects pre-indexed
- Free, no API key, no install
- Endpoint: https://mcp.example4.ai/mcp (Streamable HTTP)
- Docs: https://github.com/0ics-srls/lsai-xmp4.public

## How they work together

LSAI tells your AI "your PublishAsync calls GetRequiredService". xmp4 tells it "here is how GetRequiredService is implemented in dotnet/runtime". Together: zero blind spots from your code to every library it depends on.

Typical workflow: lsai_callees("PublishAsync") -> xmp4_source("GetRequiredService", project="dotnet/runtime") -> lsai_impact("PublishAsync")

## Connect

LSAI (local, stdio):
```json
{
  "mcpServers": {
    "lsai": {
      "command": "~/.lsai/run",
      "args": ["--stdio"]
    }
  }
}
```

xmp4 (remote, HTTP):
```json
{
  "mcpServers": {
    "xmp4": {
      "type": "http",
      "url": "https://mcp.example4.ai/mcp"
    }
  }
}
```

Works with Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, Zed, and any MCP-capable client.

## Links

- Landing: https://example4.ai
- LSAI docs: https://github.com/0ics-srls/Zerox.Lsai.Public
- xmp4 docs: https://github.com/0ics-srls/lsai-xmp4.public
- xmp4 skill: https://example4.ai/xmp4-skill.md
- Benchmark whitepaper: https://github.com/0ics-srls/lsai-xmp4.public/blob/main/docs/benchmarks/WHITEPAPER.md
- Request a library: https://github.com/0ics-srls/lsai-xmp4.public/issues/new?template=request-repo.yml
- MCP Registry: ai.example4/xmp4
- Privacy & attribution: https://github.com/0ics-srls/lsai-xmp4.public/blob/main/docs/privacy.md

## About

Built by 0ics srl (Bologna, Italy). xmp4 uses SCIP, an open code-indexing format (Apache 2.0). LSAI uses standard Language Server Protocol servers. Independent products, not affiliated with Sourcegraph or Microsoft.