mcp-standards

MCP Standards provides a personal memory module for Claude that learns and remembers your preferences automatically. It stores tool usage preferences, workflow patterns, code style tendencies, and project conventions in a local vector store (AgentDB) so Claude can apply them across sessions without manual reconfiguration. The server exposes a simple set of memory tools accessible via the MCP protocol: remember to store a preference, recall to search for remembered items, list_categories to see what domains are being tracked, memory_stats to review storage and performance, and generate_ai_standards to create a minimal Claude.md from a project configuration. To use it with Claude Desktop, configure the MCP server entry to point Claude to run the Python module that serves the standards memory, and Claude will automatically apply remembered preferences in future sessions.

Core tools available:

• remember({ content, category }): store a remembered preference (e.g., content like "use uv not pip", category like "python").
• recall({ query, top_k }): search stored preferences by intent or keywords.
• list_categories(): list the categories of remembered preferences (e.g., python, javascript, docker).
• memory_stats(): view basic statistics about the memory store.
• generate_ai_standards({ project_path, formats }): generate a CLAUDE.md or other standards docs from your project configuration.

Example flows:

• You say: "Use uv not pip". Claude stores this as a Python tool preference. In a later session, when you say "Install anything", Claude can automatically choose uv where appropriate.
• You can query: recall({ query: "package manager" }) to see related remembered preferences and quickly apply them in new tasks.

This MCP server is designed for zero-config, private, on-device memory with fast semantic search, helping Claude adapt to your personal workflows over time.

Prerequisites:

• Node.js and npm (for the project setup script and packaging). Typical requirement: Node.js v14+ and npm installed.
• Python 3.10+ for the MCP server module (as indicated by the project).
• Claude Desktop (for integration with the MCP server via the Claude config).

Install steps:

1. Clone the repository and navigate into it:

git clone https://github.com/airmcp-com/mcp-standards.git
cd mcp-standards

2. Install Node dependencies and prepare AgentDB setup (as described in Quick Start):

npm install
npm run setup

3. Ensure Python environment is ready and install any Python dependencies if required by the server module (virtualenv recommended):

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt 2>/dev/null || true

4. Configure Claude Desktop to point to the MCP server (see configuration in the README) and start Claude Desktop. The MCP server runs via uv and the Python module:

uv run --directory /ABSOLUTE/PATH/TO/mcp-standards python -m mcp_standards.server_simple

5. Restart Claude Desktop and test memory interactions:

• Remember a preference, e.g., "use uv not pip".
• Ask Claude what it remembers, e.g., "What do you remember?".

Notes:

• The project is archived and provided for reference; it runs locally and stores data under your home directory.
• The AgentDB path is typically under ~/.mcp-standards/agentdb.

Tips and caveats:

• This MCP server is designed for local, private memory storage. No cloud dependencies are used.
• The Claude Desktop integration relies on the MCP protocol; ensure that the path in the Claude config points to the mcp_standards.server_simple module.
• If you encounter issues with remembering, verify that the server is running and that Claude corrections use clear phrases like "use X not Y" to trigger auto-detection.
• AgentDB is used for fast semantic search; ensure there is enough disk space for the vector store if you accumulate many preferences.
• The project includes a few docs under docs/ and a simple CLI/test harness; these are useful for validation and exploration, but some sections may reference legacy paths since the project is archived.