Home Memory MCP is a free, standalone, open-source MCP server that gives your AI assistant structured, persistent knowledge about your home: rooms, devices, pipes, cables, and the things you own. This repository contains that server.
It works with the ChatGPT desktop app, Claude Desktop, Claude Code, Codex CLI, and other MCP-compatible AI apps. You talk naturally; the AI reads and writes your home data through Home Memory's tools.
Document what exists, plan what's coming, and keep track of what's been removed.
MCP server that lets your AI assistant remember everything about your home.
スター
77
ツール
—
稼働率
—
Your assistant is the interface for this MCP workflow. Your home data stays in a single file on your machine; depending on your assistant, information it reads may be sent to its AI provider.
Looking for a Windows workspace with linked photos and documents, reusable product information, and maintenance work orders? Home Memory Studio includes those features and its own expanded MCP. It works independently, with no separate Home Memory MCP installation required. Studio is proprietary software and free during Early Access. Its Windows download is available on the Studio page.
In this fictional home, two empty conduits from the basement utility room to the east facade are already documented. The air conditioner has just been ordered, with installation still ahead.
1. Record the plan with the order confirmation:
We've ordered this air conditioner for the living and dining room. The outdoor unit will go on the east facade. Please document the planned installation.
The assistant records the Daikin FTXM35A indoor unit and RXM35A outdoor unit at their intended locations, both marked as planned.
2. Start a new conversation:
Which AC did we choose, and do we have any spare conduits from the utility room to its planned outdoor location? Include their routes and recorded diameters.
The assistant retrieves the selected models and the two free 20 m conduits, with recorded nominal outside diameters of 25 and 40 mm. Both follow the basement ceiling and east service riser to the facade. The new plan and the existing house record come together without copying the old conversation. These are documented routes to discuss with an installer, not a guarantee of suitability.
This example uses the standalone MCP. The Studio examples show a later stage of the same fictional home, with the AC installed.
What you can do
Document anything by just talking:
"I have a Daikin Altherma heat pump in the utility room."
Your AI finds the right category, resolves the location, and creates the element. No manual data entry.
"My car is a 2023 Toyota Corolla Hybrid, next inspection is due in March."
Not just building infrastructure: vehicles, tools, appliances, valuables, anything that belongs to you.
Ask questions about your home:
"What's in the basement?" · "Show me all planned purchases." · "Where is my washing machine?"
Upload a photo and let your AI identify it:
(attach a photo of a device) "What is this? Add it to the utility room."
Vision-capable AIs recognize the device and create the element via MCP.
Read an invoice and extract devices:
(attach a PDF invoice) "Extract the installed devices and add them to my home."
Your AI client reads the attachment and records the relevant details. The standalone MCP does not store the uploaded photo or PDF as a linked file; Studio provides linked images and documents.
Track lines between elements:
"The circuit breaker panel feeds the kitchen outlet via NYM-J 3x1.5."
Cable routes, pipe runs, and duct paths, all documented as lines between elements.
Plan renovations:
"We're planning a PV system on the roof." · "The old oil heater was removed last year."
Track what's planned, what exists, and what's been removed.
Quick Start (Windows)
This installs the standalone Home Memory MCP. For Studio and its included MCP, use the Studio getting started guide instead.
The release ZIP is self-contained. No .NET or Firebird installation is required. You need one of the local AI clients listed below.
Set up with your local AI
Already use Codex or Claude Code locally on your Windows PC? It can help install and configure Home Memory. Paste this prompt:
Set up Home Memory from https://github.com/impactjo/home-memory on this Windows PC. Follow its README and Setup Guide, use the latest official release ZIP, and register it with the local AI client I am currently using. First check whether Home Memory is already installed or registered. Preserve existing client configuration and any existing Home Memory database, and update an existing registration instead of creating a duplicate. If the downloaded release contains skills\home-memory-documentation and this client supports Agent Skills, install that directory as a user-level skill. Make sure any installed skill files come from the downloaded release, not from an existing development copy. If the release does not contain the directory, complete the MCP setup without it and report that the optional documentation skill was not installed. Ask for explicit confirmation before replacing an existing installation or migrating a database. Verify the MCP connection after registration. Tell me whether the server and documentation skill were installed, and whether I need to restart the client or start a new session.
The MCP server gives your AI access to the local Home Memory database. The optional documentation skill adds rules for structuring that documentation. If the release ZIP contains skills\home-memory-documentation, that directory can be installed as a user-level skill. Clients without Agent Skills, or older releases without that directory, can still use the MCP server.
To install the skill separately, ask a client that supports Agent Skills to install skills/home-memory-documentation from this repository as a user-level skill.
After registration, continue with 3. Try it to verify the connection.
Prefer to set it up yourself? Continue with 1. Download & Extract below.
Choose one of the following clients and use a local session on the computer where you extracted Home Memory. A cloud session does not automatically have access to this installation. For phone access to the same computer, see 4. Use it from your phone below.
Click your user profile and select Settings
In the sidebar, select Plugins
Open the MCPs tab
Click Add server
Name:home-memory
Choose transport STDIO
Command:C:\HomeMemory\HomeMemoryMCP.exe (adjust if you extracted it elsewhere)
The desktop app, Codex CLI, and IDE extension share MCP configuration on the same host, so register Home Memory only once. See the official MCP guide for current client instructions.
Open Claude Desktop
Click the Claude menu → Settings → Developer → Edit Config
This opens the config folder with claude_desktop_config.json selected. Open it in any text editor
If you already have other MCP servers configured, add the "home-memory" entry next to them inside the existing "mcpServers" block.
Save the file and restart Claude Desktop
Home Memory is available in the Chat tab and in local Code tab sessions. The Code tab is recommended for multi-step, agentic workflows that use several MCP tools.
If the same server name is also configured in Claude Code, local Code tab sessions use the Desktop definition. The standalone Claude Code CLI needs its own registration below. See Claude's configuration sharing rules.
claude mcp add home-memory --scope user -- "C:\HomeMemory\HomeMemoryMCP.exe"
Check the connection with /mcp in Claude Code. If Home Memory is not available yet, start a new session.
3. Try it
On first launch, Home Memory automatically creates a local database with over 100 categories and an initial example structure (floors, rooms, garage, outdoor areas). No setup wizard needed.
First, verify that your AI can use Home Memory:
"Use the Home Memory MCP tool get_structure_overview to show me the structure currently stored in Home Memory. If the tool is unavailable, tell me instead of guessing."
You should see the initial example structure, including a ground floor, upper floor, basement, and several rooms. This confirms that your client can discover and call Home Memory's tools.
If Home Memory is unavailable or reports an error, see Troubleshooting.
Once this works, you normally do not need to mention Home Memory or individual tool names. Continue naturally:
"I have a Bosch washing machine in the basement."
Your AI creates the element, finds the right category, and places it in the basement, all in one step. Ask "What's in the basement?" to verify.
"We're planning to install a heat pump in the utility room."
Creates a planned element, so you can track what exists and what's coming.
If it works, setup is complete. From here, you can work with your AI in natural language to adapt the example structure, add appliances, furniture, or tools, and capture details from photos or invoices.
4. Use it from your phone (optional)
Home Memory runs locally, but you can use it from your phone through remote access in Claude Code or the ChatGPT desktop app. In both cases your Home Memory database and MCP server stay on your machine; the phone is just a remote window into a session running on that host.
Handy whenever you are away from your computer and want to look something up or capture it on the spot, whether at the fuse box, out in the garden, or snapping a photo of a new appliance.
Claude Code: start a local session, enable Remote Control, then continue it from the Claude mobile app or claude.ai/code. Requires a signed-in Claude account.
ChatGPT desktop app: start Remote setup on your host machine, then pair your phone by scanning the displayed QR code with the ChatGPT mobile app.
Build from Source (advanced)
See Building from Source in the Setup Guide for prerequisites, build commands, and Firebird configuration.
How it works
You ──── AI Assistant ──── Home Memory MCP ──── Local Database
(natural language) (24 tools) (Firebird, single file)
Home Memory implements the Model Context Protocol (MCP), an open standard that lets AI assistants use external tools. When you talk to your AI about your home, it calls Home Memory's tools behind the scenes to read, create, update, and search your home data.
Your data stays local. The database is a single file on your machine. Nothing is sent anywhere except to the AI you're already talking to.
Covers every domain, fully customizable to your needs
Electrical (circuits, PV, wallbox, home automation) · HVAC · Plumbing · IT & Communications · Security (alarm, fire, surveillance) · Building Materials · Landscaping (garden, pool, irrigation) · Household (appliances, furniture, valuables) · Vehicles · Tools · Health · Sports & Leisure
Every category, every element, every status is editable. Rename, add, or remove anything to match how you think about your home.
Smart defaults
Over 100 categories organized by trade and domain, from circuit breakers to garden sprinklers to vehicles
Default house structure with floors, rooms, garage, and outdoor areas, customizable by talking to your AI
Auto-setup on first run, no manual database creation needed
Flexible naming: your AI can use "Ground Floor" or "GF", the server resolves both
Compatibility
Client
Status
ChatGPT desktop app (Codex)
Tested
Claude Desktop (Chat tab + Code tab)
Tested
Claude Code (CLI)
Tested
Codex CLI (OpenAI)
Tested
Any MCP-compatible client
Should work (stdio transport)
The release ZIP is a self-contained Windows build with all dependencies included (no .NET or Firebird installation required). On macOS and Linux, you can build from source with .NET 10 and Firebird 3; see the Setup Guide for details.
Stdio is the default and works for almost everyone. HTTP mode lets you run Home Memory as a persistent service on a NAS, homelab box, or another machine on your LAN. See docs/HTTP-MODE.md for setup, database locking, LAN access, and HTTP security requirements.
Firebird Embedded: local database engine, single-file storage
Raw SQL with recursive CTEs, no ORM overhead, transparent and auditable
Contributing
Home Memory continues to evolve, and feedback is welcome. Right now, the best way to contribute is:
Open an issue for bug reports, feature ideas, or questions
Share your use case: how are you using Home Memory? What's missing?
Spread the word if you find it useful
If you'd like to build and explore the code locally, see the Setup Guide.
Background
Home Memory builds on a proven data model from Smartconstruct, a desktop application for documenting physical assets, refined through real-world use in residential construction projects.
