atlas-mcp-code
v0.1.0
Published
Repo intelligence for coding agents. Search files, find symbol definitions and references, analyze dependencies, diff commits, understand any codebase. Part of Atlas — infrastructure for AI agents.
Maintainers
Readme
atlas-mcp-code
Repo intelligence for coding agents. A Model Context Protocol server that gives Claude, Cursor, Windsurf and any AI coding agent the ability to search, navigate, and understand any local codebase — fast.
Part of Atlas — infrastructure for AI agents.
Tools
list_files— List all source files grouped by language, respects .gitignoresearch_code— Fast regex search with context lines and glob filteringread_file— Read a file (optionally with line range)file_outline— Extract imports, functions, classes, types from any source file (TS/JS/Py/Go/Rust/Java)find_symbol— Find where a symbol is defined across the repofind_references— Find every place a symbol is usedfile_stats— Repo-wide statistics: language breakdown, line counts, largest files
Why
Claude, Cursor, and Windsurf all have some code intelligence built in. It's different for each. It's rarely consistent. And when you ask "find every caller of this function," each tool has a different level of success.
Atlas Code gives you a single, portable, zero-config code intelligence layer that works identically across every MCP client. Drop it in once — every agent gets the same superpowers.
Install
{
"mcpServers": {
"atlas-code": {
"command": "npx",
"args": ["-y", "atlas-mcp-code"]
}
}
}No configuration needed. The server defaults to the current working directory of your MCP client. You can also pass a repoPath argument to every tool to target a different repo.
Design choices
- No AST parser dependency. Regex-based outline extraction is "good enough" for 95% of agent use cases, and it's 100× faster. If you need true semantic analysis, pair this with a language server.
- Respects .gitignore. Skips
node_modules,dist,.next,__pycache__, and all your custom ignore patterns automatically. - Binary-aware. Skips images, archives, fonts, and every other non-text extension by default.
- 5MB file cap. Files over 5MB are skipped to prevent accidental inclusion of minified bundles or data files.
Supported languages for outline extraction
TypeScript, JavaScript, JSX, TSX, Python, Go, Rust, Java, Kotlin.
Other languages still work with search_code, find_symbol, find_references, and file_stats — only the structured outline is language-specific.
License
MIT
