文档搜索工具

Repo Docs MCP: Secure & Flexible Documentation Search

Repo Docs MCP is a powerful tool designed to bridge the gap between large-scale AI models (like Antigravity) and technical documentation. It allows you to download, index, and search through any repository's documentation (Markdown/MDX files) using a minimalistic but highly efficient search engine.

🚀 Business Value & Use Cases

In modern AI-driven development, access to high-quality documentation is critical. While online services like Context7 are excellent for public projects, businesses often face challenges with:

1. Proprietary Documentation: Internal company docs, private SDKs, and enterprise knowledge bases cannot be shared with public indexing services due to privacy and security policies.
2. Air-Gapped Environments: Some development workflows require offline access to documentation.
3. Context Precision: LLMs work best when provided with the most relevant snippets rather than overwhelming amounts of irrelevant text.

Repo Docs MCP solves these by providing:

• Local Control: You can run the indexing and search entirely on your own infrastructure.
• Privacy First: Support for local ZIP files ensures that sensitive project documentation never leaves your machine.
• Fast Iteration: In-memory caching ensures that repetitive queries across a repository are instantaneous.

🛠️ Key Components

1. repo_docs.py (The MCP Server)

The primary interface for AI agents like Antigravity. It exposes the search_repo_docs tool, which can download a repository ZIP from GitHub (or any URL), index its documentation, and return the most relevant snippets based on a query.

2. search.py (The Standalone Script)

A flexible script for local usage. It's ideal for:

• Manual testing of the search engine.
• Searching through documentation on your local PC.
• Integrating with custom workflows where a full MCP server isn't required.

🔧 Practical Installation

Prerequisites

• Python 3.13+
• uv (recommended package manager)

Setup

Clone the repository and install dependencies:

uv sync

Running the MCP Tool

To use it with Antigravity or other MCP-compatible clients:

uv run repo_docs.py

Running the Standalone Search

To test indexing and searching locally:

uv run search.py

🤖 Usage with Antigravity

Update your Antigravity config in the mcp_config.json file:

{
    "mcpServers":  {
                       "repo_docs":  {
                                         "args":  [
                                                      "run",
                                                      "-q",
                                                      "--directory",
                                                      "C:\\tmp\\fastmcp-jina-minsearch-clone-of-context7",
                                                      "repo_docs.py"
                                                  ],
                                         "env":  {
                                                     "PYTHONUNBUFFERED":  "1"
                                                 },
                                         "command":  "C:\\Users\\d_local\\.local\\bin\\uv.exe"
                                     }
                    }
}

Later when working with Antigravity, you can simply ask:

"Search the FastMCP documentation for 'how to implement a tool' using the Repo Docs MCP."

The agent will then:

1. Identify the repository ZIP URL.
2. Use the search_repo_docs tool to index the files.
3. Retrieve the top 5 most relevant Markdown snippets.
4. Use that context to provide an accurate, document-backed response.

🔒 Security & Privacy Notice

Unlike cloud-based documentation searchers, Repo Docs MCP performs all indexing in memory. This makes it an ideal choice for enterprises that need to maintain strict data residency while still empowering their developers with AI-assisted coding.