TusharKarkera22/RepoMap-AI
Token-efficient codebase maps for AI IDEs. Parses Python, TypeScript & more with tree-sitter, builds a ranked dependency graph, and exposes it via MCP to Cursor, VS Code, and Claude Desktop.
Platform-specific configuration:
{
"mcpServers": {
"RepoMap-AI": {
"command": "npx",
"args": [
"-y",
"RepoMap-AI"
]
}
}
}Add the config above to .claude/settings.json under the mcpServers key.
[](https://pypi.org/project/repomap-ai/) [](https://www.npmjs.com/package/repomap-ai) [](https://python.org) [](LICENSE)
---
You've felt this: you're using Cursor, Claude, or Copilot on a real codebase. You ask it to refactor something. It confidently touches the wrong file, misses a dependency three folders away, or invents a function that already exists somewhere else. You spend more time correcting the AI than writing code.
The root cause: your AI doesn't know your codebase. It only sees what fits in its context window — usually a few open files. The moment your project grows beyond a handful of files, the AI starts guessing.
RepoMap fixes this by giving your AI a structured map of your entire codebase — compressed into ~1000 tokens. It's not raw source code. It's a ranked, dependency-aware index of every function, class, and their relationships, with the most important ones surfaced first.
| Without RepoMap | With RepoMap | |-----------------|--------------| | AI sees 2–3 open files | AI knows the structure of the entire repo | | Hallucinates function names | References real symbols and their signatures | | Misses cross-file dependencies | Understands what calls what, what imports what | | Needs constant copy-pasting of context | MCP server feeds context automatically on demand | | Gets confused on large codebases | Scales to 10k+ symbols via PageRank ranking |
1. Install RepoMap → pip install repomap-ai
2. Set up MCP in your IDE → repomap init . (auto-configures Cursor, VS Code, Claude Desktop)
3. Open your AI assistant → Ask anything about yLoading reviews...