ra

The ra-mcp MCP server provides access to a searchable, browsable interface for transcribed historical documents from the Swedish National Archives. It exposes a CLI and an HTTP-like MCP endpoint workflow that lets clients perform full-text search across millions of AI-transcribed pages, view complete page transcriptions, leverage handwritten text recognition, and browse documents interactively. The server also includes a CLI for direct querying (e.g., ra search, ra browse, ra tui) and can be used through streamable HTTP via an MCP client as well as integrated into Claude-style workflows with an mcp endpoint. To use it, install the ra-mcp package and run the MCP server locally, then connect your MCP client to the provided endpoint or use the built-in CLI commands to search, browse pages, or launch the interactive terminal browser. Tools available include full-text search across transcriptions, page-level browsing, text recognition integration, an interactive TUI, and documentation resources for architecture and deployment.

Prerequisites:

• Python 3.13 or newer
• Internet access to fetch packages

Installation steps:

1. Install the uvx runner (required to run UV-based MCP servers):
   • npm/yarn not required; use the uvx tool via Python ecosystem (as shown by the uv syntax in the README).
   • Install or upgrade the uvx runner if you don’t have it: uv pip install --upgrade uvx
2. Install the ra-mcp package with uv:
   • uv pip install ra-mcp
3. Run the MCP server:
   • uvx ra-mcp
   • This starts the ra-mcp MCP server so clients can connect to the configured endpoint (the exact URL may be provided by your environment or default to localhost).

Notes:

• If you are deploying in a container or a cloud environment, ensure networking allows access to the MCP endpoint and that any required environment variables are configured as documented by the ra-mcp project.

Tips and considerations:

• The server exposes a CLI (ra search, ra browse, ra tui) in addition to MCP-compatible endpoints. Use the CLI for quick local tests and validation.
• Ensure your environment has Python 3.13+ and that uvx is installed so you can run the MCP server via uvx ra-mcp.
• If you plan to expose the MCP endpoint publicly, consider securing access and monitoring usage, especially given the size of the transcribed documents set.
• For troubleshooting, consult the ra-mcp documentation at ai-riksarkivet.github.io/ra-mcp and verify that your installation matches the Quick Start guidance in the README.