# ragieai/mcp-server [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/ragieai/ragie-mcp-server  
**GitHub Stars:** 90  
**npm Downloads (last month):** 195  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/ragieai-mcp-server

## Description
Retrieve context from your Ragie (RAG) knowledge base connected to integrations like Google Drive, Notion, JIRA and more.

## Tools
Capabilities this server exposes over MCP:

- **retrieve** — Look up information in the Knowledge Base. Use this tool when you need to:
 - Find relevant documents or information on specific topics
 - Retrieve company policies, procedures, or guidelines
 - Access product specifications or technical documentation
 - Get contextual information to answer company-specific questions
 - Find historical data or information about projects

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

```json
"mcpServers": {
  "mcp-server": {
    "command": "npx",
    "args": ["-y","@ragieai/mcp-server"],
    "env": {
      "RAGIE_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `RAGIE_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What ragieai/mcp-server does

The ragieai/mcp-server MCP server gives an MCP client access to retrieval from a Ragie knowledge base. It is intended for applications such as Claude Desktop and Cursor, where an agent can search indexed organizational information during a conversation.

The repository exposes one MCP tool, `retrieve`. That tool accepts a text query and returns matching content chunks from the knowledge base. The available search controls let the caller set a maximum result count, request reranking, and optionally favor newer information.

The knowledge base may contain information drawn from connected systems such as Google Drive, Notion, and JIRA, but this repository does not provide separate MCP tools for those systems. Retrieval is performed through Ragie.

## How it works

The process communicates with an MCP client over standard input/output. When started, it reads MCP protocol messages from stdin and writes responses to stdout. A Ragie API key authorizes requests to the Ragie service.

The `retrieve` tool requires a `query` string. `topK` controls the maximum number of returned results and defaults to 8. `rerank` defaults to true and controls whether the search attempts to keep the most relevant information. `recencyBias` defaults to false and controls whether newer material receives preference.

The ragieai/mcp-server MCP server can also receive a partition identifier at startup. This allows queries to target a specified Ragie partition. The tool description can be replaced with custom text using a command-line option.

This repository is no longer required for new Ragie MCP deployments. The README directs users to Ragie’s native streamable HTTP MCP server instead. Existing users evaluating this package should account for that status before adopting it.

## Setup and configuration

Node.js 18 or newer and a Ragie API key are required. The only required environment variable is `RAGIE_API_KEY`. The package can be launched with `npx`, for example:

```bash
RAGIE_API_KEY=your_api_key npx @ragieai/mcp-server
```

Use `--partition` or `-p` to provide a Ragie partition ID. Use `--description` or `-d` to replace the default description shown for the tool. Both options can be supplied together.

Cursor can load the server from a project-level or user-level `mcp.json` file, or through a shell script registered in its MCP settings. Claude Desktop uses `claude_desktop_config.json`; after changing that file, restart the application so the server becomes available.

## Tools and capabilities

The ragieai/mcp-server MCP server provides:

- `retrieve`: searches the Ragie knowledge base and returns matching text chunks.
- `query`: supplies the search text for `retrieve`.
- `topK`: limits the number of results, with a default of 8.
- `rerank`: enables or disables relevance reranking, enabled by default.
- `recencyBias`: optionally favors more recent information, disabled by default.
- Partition targeting through the `--partition` startup option.

## Limitations and notes

Only the `retrieve` tool is documented. The server does not expose documented tools for adding, editing, or deleting knowledge-base content, and it does not provide separate tools for Google Drive, Notion, or JIRA.

The package uses stdio rather than serving MCP over HTTP. The README marks the project as no longer needed because Ragie now supports MCP natively with a streamable HTTP server. The documented implementation is written in TypeScript and uses the MCP SDK, the Ragie client library, and Zod for runtime validation.

The project is released under the MIT License.

_Full upstream README: https://allmcps.com/mcp/ragieai-mcp-server/readme_