Turn any github repo to MCP server, and chat with code or docs
Configuration
View docs{
"mcpServers": {
"probelabs-docs-mcp": {
"command": "npx",
"args": [
"-y",
"@probelabs/docs-mcp@latest",
"--gitUrl",
"https://github.com/your-org/your-repo",
"--toolName",
"search_docs",
"--toolDescription",
"Search documentation"
]
}
}
}You can turn any documentation or codebase into a searchable MCP server powered by Probe, and run it locally or in your CI. This lets AI assistants query your content naturally, helping teammates find information faster whether you’re working with docs in a local folder or a Git repository.
Set up a server that exposes a search tool to your AI assistant. You typically run the MCP server through an MCP client (such as Claude Desktop or your own client) and point the client at the server via a configured tool. The server reads content from a chosen source (a local directory or a Git repository) and makes it searchable with natural language queries. You can customize the tool name and description so the AI sees a friendly interface (for example, search_my_project_docs or search_tyk_docs). You can also target pre-built packages that bundle specific content.
Prerequisites you need before starting: Node.js installed (recommended version 18 or newer) and npm available in your system PATH. If you will pull content from a Git repository, ensure git is installed and accessible.
1. Create a default configuration file in your project root to define where the content comes from and how the MCP tool should be named. For example, you can point to a local docs directory or to a Git repository.
2. Use npm to run the MCP server via npx, which downloads and executes the package in place. You can point to a Git URL and customize the tool name and description.
// Example 1: Dynamically search a Git repository from Claude Desktop
{
"mcpServers": {
"docs_search": {
"command": "npx",
"args": [
"-y",
"@probelabs/docs-mcp@latest",
"--gitUrl",
"https://github.com/your-org/your-repo",
"--toolName",
"search_docs",
"--toolDescription",
"Search documentation"
]
}
}
}Configure your MCP client to launch the server using the exact command and arguments shown in your chosen setup. The tool exposes a search capability that your AI assistant can invoke with a natural language query. Example tool names include search_docs, search_tyk_docs, or search_tyk_official_docs.
Create a docs-mcp.config.json file in your project to define the default content source and tool details. You can switch between a local directory and a Git repository, and you can set a default tool name and description. Runtime CLI arguments or environment variables can override these defaults.
Key options you may configure include the content source (includeDir or gitUrl), the toolName and toolDescription, ignorePatterns to skip certain files, and autoUpdateInterval if you want automatic Git updates.
// Example 1: Local directory
{
"includeDir": "/Users/username/projects/my-project/docs",
"toolName": "search_my_project_docs",
"toolDescription": "Search the documentation for My Project.",
"ignorePatterns": [
"node_modules",
".git",
"build",
"*.log"
]
}// Example 2: Git repository
{
"gitUrl": "https://github.com/your-org/your-codebase.git",
"gitRef": "develop",
"autoUpdateInterval": 15,
"toolName": "search_codebase",
"toolDescription": "Search the main company codebase.",
"ignorePatterns": [
"*.test.js",
"dist/",
"__snapshots__"
]
}If you plan to publish your own pre-built MCP server, you can bundle the content into a package and publish it for others to install with a simple npx command. The configuration file remains the starting point for defaults, while runtime arguments can customize behavior.
Search documentation using the probe search engine.
Search Tyk API Management Documentation.
Search Tyk official documentation.