# LLMScribe [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/AMRITO-KUNDU/llmscribe  
**GitHub Stars:** 13  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/llmscribe

## Description
Deterministic repo map, search and file reader for AI agents.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "llmscribe": {
    "command": "uvx",
    "args": ["llmscribe"]
  }
}
```

## Documentation

## What LLMScribe MCP server does

The LLMScribe MCP server gives MCP-compatible agents a structured way to inspect code repositories. Its operations cover directory mapping, deterministic search, file retrieval, multi-file reading, dependency relationships, project snapshots, and Git changes. Outputs are machine-readable, and the underlying behavior does not use an LLM, embeddings, or inferred relevance ranking.

Repositories can be local project directories or public GitHub repositories. Public GitHub access works without cloning the repository, while local access is handled by the command-line server running on the developer's machine. The tool set supports a progressive inspection workflow: map the repository, search for relevant paths or content, then read only the files needed for the task.

## How the LLMScribe MCP server works

The server exposes these MCP tools:

- `project_map` creates a directory tree without including file contents.
- `project_overview` produces a broader snapshot with the structure, supported text-file contents, and project information.
- `search` checks file paths, names, contents, and individual lines. Results include paths, line numbers, matching lines, matched text, and match type.
- `read` returns one file after validating its path.
- `read_many` retrieves several files in one request and can return partial results when one file causes a problem.
- `project_dependencies` reports local, external, and unresolved dependencies, along with files that depend on the selected file.
- `project_diff` exposes Git status and unified diffs, including staged, unstaged, and selected commit changes.

Search results are ordered by path and line and capped at 1,000 matches. Binary and oversized files are skipped. File reads prevent traversal outside the selected project, and local `path` and GitHub `repo` inputs are mutually exclusive.

## Setup and configuration

Install the package from PyPI with `pip install llmscribe`. A local MCP configuration can launch the `llmscribe-mcp` executable over stdio:

```json
{
  "mcpServers": {
    "llmscribe": {
      "command": "llmscribe-mcp"
    }
  }
}
```

The Python module is an alternative launcher using `python -m llmscribe.mcp`. The hosted endpoint is available at `https://llmscribe.onrender.com/mcp` for clients that support remote MCP connectors. No API key is required. The hosted service is intended for experimentation; use a self-hosted local instance for private repositories or production workloads.

## Tools and capabilities

The LLMScribe MCP server is useful for agents that need predictable repository context. Mapping provides an inexpensive first look at project layout. Search can locate symbols or text before a file is read, and `read_many` can gather a known set of related files. Dependency inspection helps identify files connected to a possible change, while Git tools provide context about current modifications. The same engine is also available through the CLI and a Python core interface, although this listing focuses on its MCP interface.

## Limitations and notes

The public GitHub mode is described for public repositories and does not require cloning. The overview operation is recommended for small and medium-sized projects; larger repositories should use the map, search, and read sequence. Search is literal and deterministic rather than intent-aware, with no hidden ranking or semantic interpretation. The README names Cursor, Claude, and Windsurf as compatible MCP clients.

_Full upstream README: https://allmcps.com/mcp/llmscribe/readme_