growi-mcp 0.1.0

MCP Server corresponding to Growi, an open-source wiki tool for streamlining information sharing and knowledge management. It executes the various processes—such as page list fetching, page reading, creation, and updates, etc.—via Growi REST API internally.

Growi MCP

Growi MCP is an MCP server designed for Growi, an open-source wiki tool that streamlines information sharing and knowledge management.

Overview

This project provides an unofficial MCP server that acts as a proxy to perform various operations such as retrieving page lists, reading and writing pages, creating and updating pages, etc. Internally, it uses the Growi REST API.

Features

• Tool List

  • get_page_list(path_or_id*, limit, offset)
    • Retrieve the list of pages under the specified page (or page ID)
  • read_page(path_or_id*)
    • Read the content of the specified page (or page ID)
  • create_page(path*, body)
    • Create a page at the specified path
  • update_page(path_or_id*, body*)
    • Update the content of the specified page (or page ID)
  • rename_page(path_or_id*, new_path*)
    • Rename or move a page (by path or page ID)
  • remove_page(path_or_id*, recursively)
    • Delete a page (by path or page ID)
  • search_pages(query*, path, limit, offset)
    • Search for pages that match the query
  • get_user_names(query*, limit, offset)
    • Retrieve user names that match the query
  • upload_attachment(page_id_or_path*, file_path*)
    • Upload and attach a file to a page (by page ID or path)
  • get_attachment_list(path_or_id*, limit, offset)
    • Retrieve the list of attachments on a page (by path or page ID)
  • get_attachment_info(attachment_id*)
    • Retrieve detailed information for a page attachment (by attachment ID)
  • download_attachment(attachment_id*, save_dir)
    • Download an attachment (by attachment ID) to a local directory
  • remove_attachment(attachment_id*)
    • Delete an attachment (by attachment ID)

• MCP Server Compliance

  • Conforms to the Model Context Protocol (MCP)
  • Exposes endpoints callable from AI tools like Cursor and Cline
  • Operates via JSON-RPC over stdio

Installation

This section explains how to install uv.

Install uv

# If uv is not installed, install it first
# On Windows (PowerShell)
Invoke-WebRequest -Uri "https://astral.sh/uv/install.ps1" -UseBasicParsing | Invoke-Expression

# On Linux/macOS (shell)
curl -LsSf https://astral.sh/uv/install.sh | sh

Usage

Start the MCP server

Two ways to start the MCP server for use with Cline/Cursor are shown below.

Clone the GitHub repository and start the MCP server

If you want to clone from GitHub and run locally, clone the Growi MCP repository with the following commands:

# HTTPS
git clone https://github.com/ka1222te/growi-mcp.git
# SSH
git clone ssh://[e-mail]/ka1222te/growi-mcp.git

After cloning, run cd growi-mcp to move into the repository directory, then resolve dependencies with:

uv sync

Next, copy .env.sample to .env and configure environment variables. Example:

GROWI_DOMAIN="http://growi.example.com"
GROWI_API_TOKEN="your_access_token_here"
# API version: "1" or "3"
GROWI_API_VERSION="3"

# Optional
# If your Growi server requires a session id, set connect.sid cookie value here
#GROWI_CONNECT_SID="your_connect_sid_here"

• GROWI_DOMAIN
  • The domain where your Growi server is hosted
• GROWI_API_TOKEN
  • API token; you can issue a token from “User Settings → API Settings”
• GROWI_API_VERSION
  • Version of the Growi REST API to use. Currently supported: "1" and "3"
• GROWI_CONNECT_SID
  • Session ID required for some features (e.g., download_attachment). Access Growi via a browser and set the value of connect.sid found in the REST API request headers.

Then, add the following configuration to the settings file (JSON) of your AI coding tool (Cline/Cursor/Claude Code, etc.):

{
  "mcpServers": {
    "growi-mcp": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/growi-mcp",
        "run",
        "growi-mcp"
      ]
    }
  }
}

Replace /path/to/growi-mcp with the installation directory of Growi MCP.

Start the MCP server directly from the GitHub repository

To run directly from the GitHub repository, add the following to the settings file (JSON) of your AI coding tool:

{
  "mcpServers": {
    "growi-mcp": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/ka1222te/growi-mcp",
        "growi-mcp"
      ],
      "env": {
        "GROWI_DOMAIN": "http://growi.example.com",
        "GROWI_API_TOKEN": "your_access_token_here",
        "GROWI_API_VERSION": "3",
        "GROWI_CONNECT_SID": "your_connect_sid_here(Optional)" 
      }
    }
  }
}

If you access the GitHub repository via SSH, use uvx --from git+ssh://github.com/ka1222te/growi-mcp growi-mcp instead. Note that you must set each corresponding environment variable in the "env" section.

Examples: Using MCP Tools

Below are examples of using Growi MCP tools.

Retrieve user information

If you prompt your AI agent with the following, it can retrieve the list of users created in Growi:

List all users that exist in the wiki.

This will execute get_user_names and present the user list in alphabetical order or another structured format.

Summarize a user’s contributions

Pick one of the users obtained above and summarize their postings with a prompt like:

Summarize what the user: <username> has been writing.

This will execute tools like search_pages and read_page, collect relevant content written by the user, and then provide an LLM-generated summary.

Disclaimer

The author assumes no responsibility for any damages arising from the use of this program.

Handling of Private Information

Growi MCP executes tool/call (Function Call) requests generated by LLMs and returns responses. If your wiki is hosted in a private environment, internal information may be revealed to the LLM. Therefore, use caution when selecting which LLM to integrate. When handling confidential information, we recommend disabling the tool’s “Auto-approve” feature so that operations require explicit user approval. This helps prevent information leakage by allowing you to reject execution when accessing private pages.

About Page Updates

If you enable “Auto-approve” for update operations, there is a risk of unintentionally modifying pages written by others. Suggested mitigations:

• Restrict the scope of editable pages
  • In your prompts or a configuration file like rules.md, describe rules such as “Do not edit pages outside your own user directory.” This guides the LLM to check page ownership and helps prevent overwriting someone else’s pages.
• Disable “Auto-approve”
  • By requiring explicit user approval for each operation, you can avoid pages becoming disorganized. Toggle this setting based on your use case.

License

This project is released under the MIT License. For details, refer to the LICENSE file.