Skip to main content

An MCP Server to store and retreive project information from memory file

Project description

Project Memory MCP

An MCP Server to store and retrieve project information from memory file. This allows AI agents (like Claude) to maintain persistent memory about projects between conversations.

Overview

Project Memory MCP provides a simple way to:

  • Store project information in Markdown format
  • Retrieve project information at the beginning of conversations
  • Update project information using patches

The memory is stored in a MEMORY.md file in each project directory.

Installation

Using uvx

This method uses uvx (from the uv Python package manager) to run the server without permanent installation:

Prerequisites

Install uvx from uv if you don't have it already.

Set up MCP Client (Claude Desktop, Cursor, etc.)

Merge the following config with your existing config file (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "project-memory": {
      "command": "uvx",
      "args": [
        "project-mem-mcp",
        "--allowed-dir", "/Users/your-username/projects",
        "--allowed-dir", "/Users/your-username/Documents/code"
      ]
    }
  }
}

Note: Replace /Users/your-username with the actual path to your own projects and code directories.

Install from Source

Prerequisites

  • Python 3.11 or higher
  • Pip package manager

Clone the repository

git clone https://github.com/your-username/project-mem-mcp.git
python -m venv venv
source venv/bin/activate
pip install -e .

Set up MCP Client (Claude Desktop, Cursor, etc.)

Merge the following config with your existing config file (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "project-memory": {
      "command": "path/to/your/venv/bin/project-mem-mcp",
      "args": [
        "--allowed-dir", "/Users/your-username/projects",
        "--allowed-dir", "/Users/your-username/Documents/code"
      ]
    }
  }
}

Note: Replace /Users/your-username with the actual path to your own projects and code directories.

Arguments

The --allowed-dir argument is used to specify the directories that the server has access to. You can use it multiple times to allow access to multiple directories. All directories inside the allowed directories are also allowed. It is optional. If not provided, the server will only have access to the home directory of the user running the server.

Usage

The MCP server is started by the client (e.g., Claude Desktop) based on the configuration you provide. You don't need to start the server manually.

Tools

Project Memory MCP provides three tools:

get_project_memory

Retrieves the entire project memory in Markdown format. Should be used at the beginning of every conversation about a project.

get_project_memory(project_path: str) -> str
  • project_path: Full path to the project directory.
  • Returns the content of the MEMORY.md file as a string.
  • Raises FileNotFoundError if the project or memory file does not exist.
  • Raises PermissionError if the project path is not in allowed directories.

set_project_memory

Sets (overwrites) the entire project memory. Use this when creating a new memory file, replacing the whole memory, or if update_project_memory fails.

set_project_memory(project_path: str, project_info: str)
  • project_path: Full path to the project directory.
  • project_info: Complete project information in Markdown format.
  • Overwrites the MEMORY.md file with the provided content.
  • Raises FileNotFoundError if the project path does not exist.
  • Raises PermissionError if the project path is not in allowed directories.

update_project_memory

Updates the project memory by applying one or more block-based patches to the memory file. This is more efficient for small changes.

update_project_memory(project_path: str, patch_content: str)
  • project_path: Full path to the project directory.

  • patch_content: Block-based patch content using SEARCH/REPLACE markers (see below).

  • Each patch block must have the following format:

    <<<<<<< SEARCH
    Text to find in the memory file
    =======
    Text to replace it with
    >>>>>>> REPLACE
    

    Multiple blocks can be included in a single request.

  • Each search text must appear exactly once in the file, otherwise an error is raised.

  • Raises FileNotFoundError if the project or memory file does not exist.

  • Raises ValueError if the patch format is invalid or the search text is not unique.

  • Raises RuntimeError if patch application fails for any reason.

Example Workflow

  1. Begin a conversation with LLM about a project
  2. LLM uses get_project_memory to retrieve project information
  3. Throughout the conversation, LLM uses update_project_memory to persist new information
  4. If the update fails, LLM can use set_project_memory instead

Claude Desktop

if you use Claude Desktop, it is best to use the project feature.

Edit the project instructions:

  • Add a line like this: "The path of the project is <project_path>"
  • If it does not always use the memory, you can add a line like this: "Always use the project memory, it is not optional"

Security Considerations

  • Memory files should never contain sensitive information
  • Project paths are validated against allowed directories
  • All file operations are restricted to allowed directories

Dependencies

  • fastmcp (>=2.2.0, <3.0.0)

License

MIT

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

project_mem_mcp-0.1.0.tar.gz (9.8 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

project_mem_mcp-0.1.0-py3-none-any.whl (8.9 kB view details)

Uploaded Python 3

File details

Details for the file project_mem_mcp-0.1.0.tar.gz.

File metadata

  • Download URL: project_mem_mcp-0.1.0.tar.gz
  • Upload date:
  • Size: 9.8 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.1.0 CPython/3.13.2

File hashes

Hashes for project_mem_mcp-0.1.0.tar.gz
Algorithm Hash digest
SHA256 09590dcbdb523a07872759d8e9ce3b34483f9a9d880e5407bc3253340a916f83
MD5 0716721ab1b72042a40d8ad99d93eff0
BLAKE2b-256 133e1e18b5e7a56a5f9f256017dc723ef84ec593cf0596356618be2a3610ea2f

See more details on using hashes here.

File details

Details for the file project_mem_mcp-0.1.0-py3-none-any.whl.

File metadata

File hashes

Hashes for project_mem_mcp-0.1.0-py3-none-any.whl
Algorithm Hash digest
SHA256 cb0e8afbb6f1298c1bdc7a58944b3e15f6f31c47bebc1f1754fe46810ccfe5e5
MD5 fa36d196ed4c457039c5c69d6a126183
BLAKE2b-256 175cf524c4690efde8cd93a2d03ef655db83a478c9664a192d73b3e681ee6173

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page